Partitioning Model Research Articles

A cosine similarity-based token subsampling method for vision transformer in cloud computing

AbstractDeploying huge deep learning applications on resource-constrained edge devices is a challenging task. Cloud-based edge computing is a promising solution. Such as model partitioning, a portion of the deep learning model is deployed on the edge device; while, the remaining portion is executed by the cloud. Leveraging the computation power of edge devices, transmission latency is reduced, and bandwidth efficiency is increased. Recently, visual transformer models, supported by large datasets, have dominated in multiple vision tasks. However, model partitioning optimization methods for visual transformers are lacking. Therefore, the paper proposes a cosine similarity-based token subsampling method for visual transformer model partitioning to improve transmission efficiency. Tokens in the same class are subsampled and only the centroid tokens are uploaded. In the cloud, all tokens are reconstructed based on interpolation indexes. Three algorithm implementations are proposed and measured on PC, Jetson NANO and edge CPU Cortex-A53. The experimental results demonstrate that the recommended algorithm implementation can be executed with low-latency of 71.24 ms, and 35.65% transmitted data is reduced with an accuracy drop of 0.46%.

Phylogenetic Relationships Plays a More Important Role than Environmental Factors in Influencing Leaf Si and Ca Stoichiometry Along the Elevation Gradient in a Chinese Subtropical Forest

Silicon (Si) and calcium (Ca), as elements abundant in the Earth’s crust, are closely related to plant growth and stress resistance and have similar roles. Understanding the stoichiometry of Si and Ca can provide more insight into the mechanical and stress resistance of plants, as well as their preferences for the absorption and use of Si and Ca. Here, we measured the content of Si and Ca in the leaves of the dominant tree species located in the Mount Wuyi National Park, with an elevation ranging from 800 m to 1700 m, in an attempt to reveal changes in the Si and Ca content and ratio in the leaves along the altitude, as well as their possible relationships with environmental factors and phylogeny. The results indicated that the leaf Si and the leaf Si/Ca decreased, while the leaf Ca increased significantly with the increase in elevation. Changes in environmental factors induced by variations in elevation affected the silicon and calcium stoichiometry characteristics of the leaves, either directly or indirectly. Specifically, the mean annual precipitation, soil available silicon, soil organic matter, and soil bulk density accounted for most of the variations in leaf silicon and calcium. The leaf silicon and calcium stoichiometry was phylogenetically conservative, suggesting more similar characteristics among closely related tree species. Structural equation modeling and variation partitioning indicated that phylogeny might be more important than environmental factors in influencing leaf Si and Ca stoichiometry. Additionally, the shared effects of environmental factors and taxonomic levels indicated changes in the forest community, and the differential responses of different functional types due to elevation variation also affected the altitudinal patterns of leaf Si and Ca stoichiometry.

The partitioning of chalcophile and siderophile elements (CSEs) between sulfide liquid and carbonated melt

Carbonated melts play a significant role in mobilizing lithophile and volatile elements in the Earth’s mantle and mantle metasomatism. However, there has been limited investigation into their potential for mobilizing chalcophile and siderophile elements(CSEs). In this study, we experimentally determine the sulfide liquid–carbonated melt partition coefficients of CSEs (DCSESul/C_melt) for a range of elements, including Co, Ni, Cu, Zn, Se, Mo, Ag, Cd, In, Sn, Re, and Pb, at 1300–1600 °C, 1.0–3.0 GPa, andoxygen fugacity (fO2) close to the graphite-CO2 fluid buffer. Furthermore, the DSul/C_melt values for lithophile elements Cr, Mn, Rb, Sr, Y, Zr, Nb, Cs, Ba, Hf, and Ta (DLithoESul/C_melt) are also determined. The obtained DCSESul/C_melt values are 34–1230 for Co, 380–75200 for Ni, 200–14900 for Cu and Ag, 0.5–28 for Zn and Mo, 42–98 for Se, 24–640 for Cd, 5–52 for In and Sn, 650–15200 for Re, and 22–2470 for Pb. The obtained DLithoESul/C_melt values are below 1–10. The variations of DCSESul/C_melt and DLithoESul/C_melt are primarily influenced by the FeOtot content in the carbonated melts. A partitioning model was developed to parameterize DCSESul/C_melt and DLithoESul/C_melt as a multi-function of pressure, temperature, composition of the carbonated melt (mainly the FeOtot content), and composition of the sulfide liquid. Our parameterization can explain the observed large variations of DCSESul/C_melt and DLithoESul/C_melt for most of the trace elements studied. Using our DCSESul/C_melt parameterization, we model the CSE and U–Th contents of low-degree partial melts of carbonated mantle peridotite and slab eclogite with sulfur concentrations ranging from 50 to 500 µg/g. The modeling results can generally explain the trace element patterns observed in natural kimberlites and carbonatites; however, the peridotite- or slab-derived carbonated melts have a low capability in mobilizing CSEs, which can extract less than 3 % of Cu, Ni, Co, Re, and Os, 3–30 % of Mo, Pb, and Se, but up to 30–50 % U and Th from the source lithology. Consequently, the influence of carbonatite metasomatism on the Cu, Ni, Co, Re, and Os systematics of the Earth’s mantle is minimal, although local enrichments of CSEs may occur when sulfides precipitate from carbonated melts. Because of the elevated concentrations of U and Th and the corresponding U/Pb and Th/Pb ratios in the carbonated melts, the mantle lithology that has undergone metasomatism by these melts can become a geochemical reservoir with high 208Pb/206Pb ratios. However, the effect of carbonatite metasomatism on Re–Os isotopic systems of the mantle is minimal due to the low Re concentrations in the carbonated melts. Accordingly, the radiogenic Pb–Os isotopic signatures of HIMU ocean island basalts cannot be explained solely by carbonatite metasomatism in the mantle.

Simulation of Small-Break Loss-of-Coolant Accident Using the RELAP5 Code with an Improved Wall Drag Partition Model for Bubbly Flow

The RELAP5 code is a computational tool designed for transient simulations of light water reactor coolant systems under hypothesized accident conditions. The original wall drag partition model in the RELAP5 code has a problem in that the bubble velocity is predicted to be faster than the water velocity in the fully developed flow in a constant-area channel. The wall drag partition model, based on the wetted perimeter concept, proves insufficient for accurately modeling bubbly flows. In this study, the wall drag partition model was modified to account for the physical motion of fluid particles. After that, the modified RELAP5 code was applied to predict the SBLOCA of a full-scale nuclear power plant. Considering the SBLOCA scenario, the behavior change in the loop seal clearing phenomenon was clearly shown in the analysis by the model change. Upon the termination of natural circulation, the loop seals were cleared, allowing the steam trapped within the system to discharge through the break. The modified model was confirmed to have an impact at this time. It mainly affected the timing and shape of the loop seal clearing and delayed the overall progress of the accident. It was observed that the flow rate of the bubbly phase decreased as the modified model accounted for wall friction during dispersed flow in the horizontal section, impacting the two-phase flow behavior at the conclusion of the natural circulation phase.

Wall heat partitioning model with bubble tracking method for nucleate boiling considering conjugate heat transfer coupled with OpenFOAM

This study aimed to predict boiling heat transfer more accurately by incorporating the bubble tracking method and conjugate heat transfer into the conventional heat partitioning model. The bubble tracking method is developed to predict boiling heat transfer by continuously simulating the size and location of individual bubbles and simulating realistic phenomena in boiling, unlike previous methods for predicting boiling heat transfer. The method considers several factors that were not previously considered, including the stochastic behavior of the boiling process, interaction between bubbles, interaction between nucleation sites, and microlayer evaporation considering the thickness and radius of the microlayer. Additionally, it was validated for pool boiling experiments. In this study, the model was improved by incorporating conjugate heat transfer for the wall temperature variation that was not considered in the previous study, and the method was also validated with the pool boiling experiment. In this validation, the temporal and spatial surface temperature variation seen in the single-bubble experiment was well simulated, and the trends in heat flux observed in the multi-bubble experiment were well predicted.

Metabolic analysis of polymeric proanthocyanidins related to red pigmentation in Camellia drupifera cv. ‘Hongrou No.1’ mesocarps

Red mesocarp, characterized as a unique pigment trait of newly identified Camellia drupifera cv. ‘Hongrou No.1’(‘HR’), is believed to act as the plant's protective shield against diverse adversities. Comprehensive metabolic profiling revealed that the ectopic deposition of polymeric insoluble proanthocyanidins (PAs) in cell walls is responsible for the “red” pigmentation of ‘HR’ mesocarps. Furthermore, structural equation modeling and variation partitioning analysis demonstrated that a molybdenum-dependent aldehyde oxidase, encoded by CdGLOX1, was induced in ‘HR’ mesocarps and deemed to be a dominant determinant of polymeric insoluble PA accumulation through the putative oxidative condensation of PA subunits. This study provides a background for an in-depth understanding of the mechanisms of unperceived pigmentation in fruits.

Factorization of rational six vertex model partition functions

We show factorization formulas for a class of partition functions of rational six vertex model. First we show factorization formulas for partition functions under triangular boundary. Further, by combining the factorization formulas with the explicit forms of the generalized domain wall boundary partition functions by Belliard-Pimenta-Slavnov, we derive factorization formulas for partition functions under trapezoid boundary which can be viewed as a generalization of triangular boundary. We also discuss an application to emptiness formation probabilities under trapezoid boundary which admit determinant representations.

Exploring How the Surface-Area-to-Volume Ratio Influences the Partitioning of Surfactants to the Air-Water Interface in Levitated Microdroplets.

Quantitative characterization of the surface of microdroplets is important to understanding and predicting numerous chemical and physical processes, such as cloud droplet formation and accelerated chemistry in microdroplets. However, it is increasingly appreciated that the surface compositions of microdroplets do not necessarily match those of macroscale solution due to their large surface-area-to-volume (SA-V) ratios and confined volumes. In this work, we explore how both droplet size and composition affect the surface composition of microdroplets by measuring the equilibrium surface tensions of levitated microdroplets containing a single surfactant. We measure the critical micelle concentrations (CMCs) for surfactants of various strengths (macroscale CMC values ranging from 0.02 to 10 mM) in microdroplets with radii ranging from 5 to 25 μm. We accurately model the surface tensions of microdroplets using an equilibrium partitioning model that only requires droplet size and adsorption parameters from macroscale measurements as inputs. Our model predicts that surfactants have an "effective CMC" in microdroplets that is always larger in value than the corresponding macroscale CMC. In some instances, the effective CMC of a surfactant in microdroplets is several orders of magnitude larger than both its macroscale CMC and its macroscale solubility limit. We present a simple expression for the effective CMC in microdroplets that depends on both the macroscale CMC of a surfactant and the SA-V ratio of the microdroplet. Ultimately, our experimental results and model can be used broadly to predict microdroplet surface compositions when investigating surface-driven accelerated chemistry in microdroplets or estimating cloud droplet activation.

Bad Local Minima Exist in the Stochastic Block Model

We study the disassortative stochastic block model with three communities, a well-studied model of graph partitioning and Bayesian inference for which detailed predictions based on the cavity method exist (Decelle et al. in Phys Rev E 84:066106, 2011). We provide strong evidence that for a part of the phase where efficient algorithms exist that approximately reconstruct the communities, inference based on maximum a posteriori (MAP) fails. In other words, we show that there exist modes of the posterior distribution that have a vanishing agreement with the ground truth. The proof is based on the analysis of a graph colouring algorithm from Achlioptas and Moore (J Comput Syst Sci 67:441–471, 2003).

Investigation of bubble interaction in a temporal and spatial heat flux partitioning model for subcooled flow boiling

CFD methodology for subcooled flow boiling is significantly affected by the heat flux partitioning model, which plays a crucial role in predicting the mass source term generated by wall boiling of the liquid phase. Although numerous models have been proposed, most research efforts have focused on the spatial dimensions of boiling heat transfer mechanisms, neglecting the temporal aspect. This study presents a heat flux partitioning model for subcooled flow boiling that considers both temporal and spatial dimensions. The model accounts for the contribution of heat flux from the superheated liquid layer during the bubble growth stage, liquid convection during the bubble wait stage within the bubble influence area in the temporal dimension, and heat flux from the overlapping area of bubble influence in the spatial dimension. An approach was developed to determine the bubble influence area by considering the bubble interaction based on the stochastic nature of nucleation sites on the heated wall, verified by the Monte Carlo method. The bubble dynamics parameters were experimentally derived to reduce the uncertainty associated with the boiling sub-models on the calculation. A comparative analysis of boiling curves and wall heat flux partitioning was conducted between the present model and the RPI model under various flow conditions. The results indicate that both models could reasonably predict the boiling curve. However, the RPI model tends to over-predict the proportion of evaporative heat flux, which is considered a weakness. Based on physical principles, the present model accurately captures the fundamental trend of wall heat flux partitioning. This research enhances the understanding of subcooled flow boiling and increases confidence in multiphase CFD methodology predictions.

Development of an ultrasound-based clinical decision rule to rule-out diverticulitis

The concern for diverticulitis often leads to the use of computed tomography (CT) scans for diagnosis. We aim to develop an ultrasound-based clinical decision rule (CDR) to confidently rule-out the disease without requiring a CT scan. We analyzed data from a prospective study of adult emergency department (ED) patients with suspected diverticulitis who underwent both bedside ultrasound (US) and CT. Patient history, physical examination, laboratory findings, and US results were used to create a CDR via a recursive partitioning model designed to prioritize sensitivity, with a loss matrix heavily penalizing false negatives. We calculated the test characteristics for this CDR (TICS-Rule) and assessed the potential reduction in CT scans and ED length of stay. Data from 149 patients (84 female; mean age 58 ± 16) were used to develop the TICS-Rule. The final model integrates US diagnosis of simple and complicated diverticulitis, along with variables of heart rate, age, history of diverticulosis, vomiting, and leukocytosis. Negative US results and a heart rate below 100 effectively excluded diverticulitis. The sensitivity increased from 54.5% (32.2–75.6) in the US alone to 100% (84.6–100%) for complicated diverticulitis in the model. The TICS-Rule missed no cases of complicated diverticulitis but one case of simple diverticulitis. The median time from ED greeting to US interpretation was 103 min (IQR 62–169), compared to 285 min (IQR 229–372) for CT. The TICS-Rule uses a combination of negative US and heart rate less thanQ1 100 to exclude diverticulitis without the need for a CT scan. Integration of the TICS-Rule offers a promising enhancement to clinical decision-making while reducing both CT use and ED length of stay.

Changing disc compositions via internal photoevaporation

The chemical evolution of protoplanetary discs is a complex process that is not fully understood. Several factors influence the final spatial distribution of atoms and molecules in the disc. One such factor is the inward drift and evaporation of volatile-rich pebbles that can enrich the inner disc with vapour. In particular, the inner disc is first enriched with evaporating water-ice, resulting in a low C/O ratio, before carbon-rich gas from the outer disc – originating from the evaporation of CO, CO2, and CH4 ice – is transported viscously inwards, elevating the C/O ratio again. However, it is unclear how internal photoevaporation – which carries away gas and opens gaps in the disc that can block inward drifting pebbles – affects the chemical composition of the disc. Our goal is to study how and to what extent internal photoevaporation and the subsequent opening of gaps influence the chemical evolution of protoplanetary discs around solar-like stars (M* = 1 M⊙), where we specifically focus on the C/O ratio and the water content. To carry out our simulations, we use a semi-analytical 1D disc model. The code chemcomp includes viscous evolution and heating, pebble growth and drift, pebble evaporation and condensation, as well as a simple chemical partitioning model for the disc. We show that internal photoevaporation plays a major role in the evolution of protoplanetary discs and their chemical composition: As photoevaporation opens a gap, inward drifting pebbles are stopped and can no longer contribute to the volatile content in the gas. In addition, volatile-rich gas from the outer disc, originating from evaporated CO, CO2, or CH4 ice, is carried away by the photoevaporative winds. Consequently, the C/O ratio in the inner disc remains low. In contrast, gaps opened by giant planets still allow the gas to pass, resulting in an elevated C/O ratio in the inner disc, similar to the evolution of viscous discs without internal photoevaporation. This opens the possibility to distinguish observationally between these two scenarios when measuring the C/O ratio, implying that we can infer the root cause of deep gap structures when observing protoplanetary discs. In the case of a clear separation of the disc by photoevaporation, we additionally find an elevated water content in the inner disc, because the water vapour and ice undergo a cycle of evaporation and recondensation, preventing the inward accretion of water onto the star, in contrast to the situation for hydrogen and helium. We conclude that it is very difficult to achieve supersolar C/O ratios in the inner parts of protoplanetary discs when taking internal photoevaporation into account. This indicates the potential importance of photoevaporation for understanding the chemical evolution of these discs and the planets forming in them.

MemoriaNova: Optimizing Memory-Aware Model Inference for Edge Computing

In recent years, deploying deep learning models on edge devices has become pervasive, driven by the increasing demand for intelligent edge computing solutions across various industries. From industrial automation to intelligent surveillance and healthcare, edge devices are being leveraged for real-time analytics and decision-making. Existing methods face two challenges when deploying machine learning models on edge devices. The first challenge is handling the execution order of operators with a simple strategy, which can lead to a potential waste of memory resources when dealing with directed acyclic graph structure models. The second challenge is that they usually process operators of a model one by one to optimize the inference latency, which may lead to the optimization problem getting trapped in local optima. We present MemoriaNova, comprising BTSearch and GenEFlow, to solve these two problems. BTSearch is a graph state backtracking algorithm with efficient pruning and hashing strategies designed to minimize memory overhead during inference and enlarge latency optimization search space. GenEFlow, based on genetic algorithms, integrates latency modeling and memory constraints to optimize distributed inference latency. This innovative approach considers a comprehensive search space for model partitioning, ensuring robust and adaptable solutions. We implement BTSearch and GenEFlow and test them on eleven deep-learning models with different structures and scales. The results show that BTSearch can reach 12% memory optimization compared with the widely used random execution strategy. At the same time, GenEFlow reduces inference latency by 33.9% in distributed systems with four-edge devices.

DeepZoning: Re-accelerate CNN Inference with Zoning Graph for Heterogeneous Edge Cluster

Parallelizing CNN inference on heterogeneous edge clusters with data parallelism has gained popularity as a way to meet real-time requirements without sacrificing model accuracy. However, existing algorithms struggle to find optimal parallel granularity for complex CNNS, the structure of which is a directed acyclic graph (DAG) rather than a chain, and the parallel dimension is inflexible. To distribute the workload of modern CNNs on heterogeneous devices is also proven as NP-hard problem. In this paper, we introduce DeepZoning , a versatile and cooperative inference framework that combines both model and data parallelism to accelerate CNN inference. DeepZoning employs two algorithms at different levels: (1) a low-level Adaptive Workload Partition algorithm that uses linear programming and takes spatial and channel dimensions into optimization during the search for feature map distribution on heterogeneous devices, and (2) a high-level Model Partition algorithm that finds the optimal model granularity and organizes complex CNNs into sequential zones to balance communication and computation during execution. Our experimental evaluations show that DeepZoning is effective, achieving up to a 3.02 × speed improvement on our experimental prototype compared to state-of-the-art algorithms.

A partitioning Monte Carlo approach for consensus tasks in crowdsourcing

The planner’s role in crowdsourcing involves determining the time to stop collecting information (i.e., timed decision, TD). Previous studies have modeled the uncertain crowdsourcing environment as Partially Observable Markov Decision Processes (POMDPs) and utilized the value of information (VOI) to balance the utilities and costs of information collection. However, there is still room for optimization in solving the TD problem within single-agent POMDPs. In this paper, we propose a partitioning Monte Carlo approach for consensus tasks based on the option-candidate (OC) model partitioning. We simplify the state representation for the OC model and introduce a multi-agent POMDP representation. We establish a correspondence between the single-agent and multi-agent models using two consistency theorems. We propose three progressively improved partitioning Monte Carlo (PMC) algorithms to solve the TD problem within the multi-agent POMDP. We conducted experiments in synthetic domains and a citizen science project, demonstrating that the proposed algorithms exhibit continuous improvements in runtime advantages and consistently outperform the SOTA single-agent Monte Carlo sampling-based algorithm while providing nearly consistent output results.

Origin of division of labor is decoupled from polymorphism in colonial animals.

Division of labor, the specialization of sometimes phenotypically divergent cell types or group members, is often associated with ecological success in eukaryotic colonial organisms. Despite its many independent evolutionary origins, how division of labor emerges remains unclear. Conventional hypotheses tend toward an "economic" model, so that biological division of labor may reflect a partitioning of preexisting tasks and morphologies into specialized colony members. Here, we present an alternative model of the origin of division of labor, which can explain the evolution of new functions within a colony. We show that in colonies of the Cretaceous aged (103-96 Ma) fossil bryozoan of the genus Wilbertopora, the first cheilostome bryozoan to evolve polymorphism, preexisting morphologies were not simply partitioned among new members, but instead expanded into novel morphospace as they lost functions, specifically feeding. This expansion occurred primarily during two pulses of heightened morphological disparity, suggesting that the evolution of polymorphism corresponded to relaxed constraints on morphology and perhaps to the exploration of novel functions. Using a simple model of physiological connections, we show that regardless of the functionality of these new colony members, all nonfeeding members could have been supported by neighboring feeding members. This suggests that geometric constraints and physiological connectedness could be prerequisites for evolving both polymorphism and division of labor in modular organisms, and that a classic partitioning model of specialization cannot be broadly applied to biological systems.

Enhancing Federated Learning in Heterogeneous Internet of Vehicles: A Collaborative Training Approach

The current Internet of Vehicles (IoV) faces significant challenges related to resource heterogeneity, which adversely impacts the convergence speed and accuracy of federated learning models. Existing studies have not adequately addressed the problem of resource-constrained vehicles that slow down the federated learning process particularly under conditions of high mobility. To tackle this issue, we propose a model partition collaborative training mechanism that decomposes training tasks for resource-constrained vehicles while retaining the original data locally. By offloading complex computational tasks to nearby service vehicles, this approach effectively accelerates the slow training speed of resource-limited vehicles. Additionally, we introduce an optimal matching method for collaborative service vehicles. By analyzing common paths and time delays, we match service vehicles with similar routes and superior performance within mobile service vehicle clusters to provide effective collaborative training services. This method maximizes training efficiency and mitigates the negative effects of vehicle mobility on collaborative training. Simulation experiments demonstrate that compared to benchmark methods, our approach reduces the impact of mobility on collaboration, achieving large improvements in the training speed and the convergence time of federated learning.

A sub-wavelength metamaterial layer for reducing underwater noise radiation of marine structures

Reducing the underwater noise radiated by marine structures is a critical issue for improving submarine stealth or limiting the anthropogenic impact of commercial ships. In this study, we propose a lightweight passive solution based on a metamaterial to efficiently reduce the noise at low frequency (< 1000 Hz) that can be transmitted through the shell plating of a marine vessel. This solution is made of a soft acoustically impervious poro-elastic foam in which inclusions are periodically positioned. Added to a plate, this metamaterial outperforms the mass law in the sub-wavelength domain when separating two air domains. This performance is due to the periodic arrangement of inclusions in the foam, creating a vibro-acoustic bandgap. The aim of this study is therefore to assess this solution where the domain of the transmitted acoustic wave is water. Firstly, numerical results on an unbounded plate are validated alongside the theoretical model of a flexible partition. When the metamaterial layer is added to the plate, it is shown that a vibro-acoustic bandgap still exists in the transmitted wavenumber domain, and that this solution is a good candidate for achieving the objectives. Experimental measurements on a finite structure are then conducted and compared to numerical results

Global Influence of Organic Aerosol Volatility on Aerosol Microphysical Processes: Composition and Number

AbstractWe present MATRIX‐VBS, a new aerosol scheme that simulates organic partitioning in an aerosol microphysics model, as part of the NASA GISS ModelE Earth System Model. MATRIX‐VBS builds on its predecessor aerosol microphysics model MATRIX (Bauer et al., 2008, https://doi.org/10.5194/acp‐8‐6003‐2008) and was developed in the box model framework (Gao et al., 2017, https://doi.org/10.5194/gmd‐10‐751‐2017). The scheme features the inclusion of organic partitioning between the gas and particle phases and the photochemical aging process using the volatility‐basis set (Donahue et al., 2006, https://doi.org/10.1021/es052297c). To assess the performance of the new model, we compared its mass concentration, number concentration, and activated number concentration to the original scheme MATRIX, and evaluated its mass concentrations in four seasons against data from the NASA Atmospheric Tomography Mission (ATom) aircraft campaign. Results from MATRIX‐VBS show that organics are transported further away from their source, and their mass concentration increases aloft and decreases at the surface compared to those in MATRIX. The mass concentration of organics at the surface agrees well with measurements, and there are discrepancies for vertical profiles aloft. In the new scheme, the global mass load of organic aerosols increased by 50%, there is also an increased number of particles at the surface and fewer activated ones in most regions. The new scheme presents advanced and more comprehensive capability in simulating aerosol processes.

Multiphase Reactions of Hydrocarbons Into an Air Quality Model With CAMx‐UNIPAR: Impacts of Humidity and NOx on Secondary Organic Aerosol Formation in the Southern USA

AbstractSecondary organic aerosol (SOA) mass in the Southern USA during winter‐spring 2022 was simulated by integrating Comprehensive Air quality Model with extensions (CAMx) with the UNIfied Partitioning‐Aerosol phase Reaction (UNIPAR) model, which predicts SOA formation via multiphase reactions of hydrocarbons. UNIPAR streamlines multiphase partitioning of oxygenated products and their heterogeneous reactions by using explicitly predicted products originating from 10 aromatics, 3 biogenics, and linear/branched alkanes (C9‐C24). UNIPAR simulations were compared with those using Secondary Organic Aerosol Partitioning (SOAP) model, which uses simple surrogate products for each precursor. Both UNIPAR and SOAP showed similar tendencies in SOA mass but slightly underpredicted against observations at given five ground sites. However, SOA compositions and their sensitivity to environmental variables (sunlight, humidity, NOx, and SO2) were different between two models. In CAMx‐UNIPAR, SOA originated predominantly from alkanes, terpenes, and isoprene, and was influenced by humidity, showing high SOA concentrations with wet‐inorganic salts, which accelerated aqueous reactions of reactive organic products. NO2 was positively correlated with biogenic SOA because elevated levels of nitrate radicals and hygroscopic nitrate aerosol effectively oxidized biogenic hydrocarbons at night and promoted SOA growth via organic heterogeneous chemistry, respectively. Anthropogenic SOA, which formed mainly via daytime oxidation with OH radicals, was weakly and negatively correlated with NO2 in cities. In CAMx‐UNIPAR, the sensitivity of SOA to aerosol acidity (neutral vs. acidic aerosol at cation/anion = 0.62) was dominated by isoprene SOA. The reduction of NOx emissions could effectively mitigate SOA burdens in the Southern USA where biogenic hydrocarbons are abundant.