这是用户在 2024-6-9 14:09 为 https://app.immersivetranslate.com/pdf-pro/dbb8d72a-12e7-4a94-8c07-d9d105769c26 保存的双语快照页面,由 沉浸式翻译 提供双语支持。了解如何保存?
2024_06_09_d15dacd90ebe03a6b32fg

Counterintuitive Oxidation of Alcohols at Air-Water Interfaces
酒精在空气-水界面上的反直觉氧化作用

Deming Xia, Jingwen Chen, Hong-Bin Xie,* Jie Zhong, and Joseph S. Francisco*
Deming Xia、Jingwen Chen、Hong-Bin Xie、* Jie Zhong 和 Joseph S. Francisco*

Cite This: J. Am. Chem. Soc. 2023, 145, 4791-4799
引用此文:J. Am.Chem.2023, 145, 4791-4799
Read Online 在线阅读
ACCESS | Metrics & More Article Recommendations
ACCESS | 指标及更多文章推荐
(s1 Supporting Information
(s1 佐证资料

Abstract 摘要

This study shows that the oxidation of alcohols can rapidly occur at air-water interfaces. It was found that methanediols orient at air-water interfaces with a atom of the group pointing toward the gaseous phase. Counterintuitively, gaseous hydroxyl radicals do not prefer to attack the exposed group but the group that forms hydrogen bonds with water molecules at the surface via a waterpromoted mechanism, leading to the formation of formic acids. Compared with gaseous oxidation, the water-promoted mechanism at the air-water interface significantly lowers free-energy barriers from to and therefore accelerates the formation of formic acids. The study unveils a previously overlooked source of environmental organic acids that are bound up with aerosol formation and water acidity.
这项研究表明,醇类的氧化可在空气-水界面迅速发生。研究发现,甲二醇 在空气-水界面上定向时, 基团的 原子指向气相。与直觉相反,气态羟基自由基并不倾向于攻击暴露在外的 基团,而是攻击 基团,后者通过水促进机制与表面的水分子形成氢键,从而形成甲酸。与气态氧化相比,空气-水界面上的水促进机制大大降低了从 的自由能垒,因此加速了甲酸的形成。这项研究揭示了以前被忽视的与气溶胶形成和水酸性有关的环境有机酸来源。

■ INTRODUCTION 引言

Organic acids (OAs) that account for of total organic carbons are ubiquitous in the environment. In the atmosphere, OAs can directly participate in aerosol formation via enhancing new particle formation and contributing to secondary organic aerosols, affecting global climate and local air quality. In the hydrosphere (e.g., cloud, dew, and sea), increasing shreds of evidence have indicated that aqueousphase acidity is strongly influenced by OAs, of which heterogeneous chemistry is also highly related to local ozone and particle formation. In the pedosphere, mineral weathering, metal detoxification, and nutrient acquisition by plant roots are associated with the participation of OAs. Therefore, OAs play important roles in the environment.
占总有机碳 的有机酸(OAs)在环境中无处不在。 在大气中,有机酸可以通过增强新粒子的形成和促成二次有机气溶胶而直接参与气溶胶的形成, 从而影响全球气候和当地空气质量。在水圈(如云层、露水和海洋)中,越来越多的证据表明,水相酸度受到 OAs 的强烈影响,其中的异质化学性质也与当地臭氧和颗粒物的形成密切相关。 在土壤圈中,矿物风化、金属解毒和植物根系获取养分都与 OAs 的参与有关。 因此,OA 在环境中发挥着重要作用。
However, environmental sources of OAs remain elusive. Existing shreds of evidence from laboratory experiments and field observations suggested that most OAs are not emitted directly from primary sources (e.g., vegetation, microorganism, and biomass burning) but are produced from multitudinous secondary sources, especially oxidation of alcohols at their positions (i.e., the carbons nearest to the group of the alcohols). For example, gaseous oxidation of methanediol , the hydrated form of formaldehyde , by hydroxyl radicals (. ) has been recently shown to be a previously underestimated source of atmospheric formic acid -the simplest and the most abundant in the atmosphere. Nevertheless, the same pathway has also been found to be unimportant in explaining the mismatch in diurnal trends of simulated and observed concentrations. The disagreement implies some missing mechanisms of alcohol oxidation in addition to the gaseous oxidation.
然而,OAs 的环境来源仍然难以捉摸。 实验室实验和实地观察所获得的现有证据表明,大多数 OAs 并不是直接从原生源(如植被、微生物和生物质燃烧)排放的,而是从多种次生源产生的,特别是醇类在其 位置(即最靠近醇类 基团的碳)的氧化。 例如,甲二醇 (甲醛 的水合形式)被羟自由基(. )气态氧化,最近已被证明是大气中甲酸 (大气中最简单、最丰富的甲酸 )以前被低估的来源。 然而,同样的途径在解释模拟和观测到的 浓度昼夜变化趋势不一致时也被认为并不重要。 这种差异意味着除了气态氧化之外,酒精氧化的某些机制也存在缺失。
Alcohols as well as many other organic tracers may be enriched at the air-water interfaces due to their "surfactantlike" structures, i.e., containing both hydrophilic and hydrophobic functional groups. Methanol and ethanol have been experimentally and computationally found to be accumulated at the airwater interfaces with adsorption-free-energies of -5.2 to -3.9 Furthermore, the water surface has been shown to affect the energies of the frontier orbitals of adsorbates (e.g., formaldehyde, the dehydrated form of ) via acting as an H-bond donor or acceptor. This could favor the reaction of free radicals with the adsorbates, triggering chain oxidation reactions. In addition, "surface preference" of at the air-water interfaces has also been reported. The above circumstantial evidence suggests that a unique and important transformation from alcohols to OAs may exist at the airwater interfaces.
醇类以及许多其他有机示踪剂由于其 "类似表面活性剂 "的结构(即同时含有亲水的 和疏水的 官能团),可能会在空气-水界面富集。 实验和计算发现,甲醇 和乙醇 在空气-水界面上积聚,吸附自由能为 -5.2 到 -3.9 此外,研究还表明,水面会影响吸附物(如:甲醛、脱水甲醛、脱水乙醇)前沿轨道的能量、 的脱水形式甲醛)。 这有利于自由基与吸附剂发生反应,引发链式氧化反应。 此外, 在空气-水界面的 "表面偏好 "也有报道。 上述间接证据表明,在空气-水界面上可能存在从醇到 OAs 的独特而重要的转化过程。
Despite difficulties in measuring interface phenomena experimentally, recent advances in machine learning make it possible to simulate interfacial reactions. In this study, a reactive machine-learning force field (MLFF) that can reach density-functional-theory-level quality with a significant decrease in computational costs was trained based on ab initio data selected from ca. configurations via an on-the-
尽管难以通过实验测量界面现象, 机器学习的最新进展使模拟界面反应成为可能。 在这项研究中,我们根据从约 个构型中选取的 ab initio 数据,训练了一个反应性机器学习力场(MLFF),它可以达到密度泛函理论级别的质量,同时显著降低计算成本。
Figure 1. Performance of trained machine-learning force field (MLFF) in predicting (A) energies ( 7,018 and 1,800 data points for the training and external validation sets, respectively), (B) forces ( and 408,300 data points for the training and external validation sets, respectively), (C) stresses ( 42,108 and 13,800 data points for the training and external validation sets, respectively), and radial distribution functions (RDFs) for (D) distances in the aqueous phase, (E) distances ( , and ( ) distances ( ) at water surfaces with experimental or initio molecular dynamics (AIMD) data as comparisons. (The white, gray, and red balls in panels represent , and atoms, respectively; the AIMD data in panels and were calculated using the functional with D4 correction in this study.)
图 1.训练有素的机器学习力场 (MLFF) 在预测 (A) 能量(训练集和外部验证集的数据点分别为 7,018 和 1,800 个)、(B) 力(训练集和外部验证集的数据点分别为 和 408,300 个)、(C) 应力(训练集和外部验证集的数据点分别为 42,108 和 13,800 个)以及 (D) 水相中 距离 的径向分布函数 (RDF) 方面的性能、(D) 水相中 距离 的径向分布函数 (RDF),(E) 水表面 距离 ( ) 和 ( ) 距离 ( ) 的径向分布函数 (RDF),以实验 初始分子动力学 (AIMD) 数据作为比较。(面板 中的白球、灰球和红球分别代表 原子;面板 中的 AIMD 数据是本研究中使用 函数和 D4 修正 计算得出的)。
fly algorithm. Based on the developed MLFF, a series of machine-learning-based molecular dynamics simulations with the assistance of ab initio molecular dynamics (AIMD), enhanced sampling techniques, and quantum chemical calculations was carried out to reveal the transformation mechanisms and kinetics of at the air-water interfaces. The heterogeneous mechanism presented here may facilitate elucidating the high levels of OAs that affect aerosol growth, cloud evolution, and water/soil acidity.
飞算法。 基于开发的MLFF,在ab initio分子动力学(AIMD)、增强采样技术和量子化学计算的辅助下,进行了一系列基于机器学习的分子动力学模拟,揭示了 在空气-水界面的转化机制和动力学。本文提出的异质机制可能有助于阐明影响气溶胶生长、云演变和水/土壤酸度的高浓度 OAs。

■ METHODS 方法

On-the-Fly Ab Initio Molecular Dynamics (AIMD) Simulation. All the AIMD simulations were performed using a recently developed SCAN functional with plane-wave basis sets cutoff in energy) on VASP 6.3. The D4 correction method was applied to account for weak dispersion interactions. The combination of and D4 (R्र ) has been recently proven to approach the accuracy of hybrid functionals for general chemical applications (especially for main group elements) with relatively low computational costs. The projector-augmented-wave pseudopotentials with nonspherical contributions were implemented to treat core electrons. The MLFFs were trained on-the-fly during the AIMD simulations based on the Bayesian linear regression in the NVT ensemble at ca. that was controlled by the Nosé-Hoover thermostat method. The time step was .
在线 Ab Initio 分子动力学(AIMD)模拟。所有的 AIMD 模拟都是在 VASP 6.3 上使用最近开发的 SCAN 函数和平面波基集 能量截止进行的。 采用了 D4 修正方法来考虑弱色散相互作用。 和 D4(R्र )的组合最近已被证明能够以相对较低的计算成本接近混合函数在一般化学应用(特别是主族元素)中的精度。 采用了具有非球面贡献的投影增强波伪势来处理核心电子。 在AIMD模拟期间,根据NVT集合中的贝叶斯线性回归对MLFF进行了即时训练。 由 Nosé-Hoover 恒温器方法控制。 时间步长为
Major hyperparameters used for the MLFF construction were optimized with the on-the-fly AIMD simulations for six independent systems (Tables S1 and S2). In total, 7017 key first-principle points were sampled using an active-learning algorithm based on almost simulation steps ( ) of the on-the-fly simulations. The trained MLFF model consisted of 4336, 5053, and 2894 basis functions for , and elements, respectively. The MLFF model was first validated on an external data set that was not used for training the model. The data set consisted of 1800 structures covering different phases (i.e., gaseous phase, aqueous phase, and air-water interfaces) and reaction statuses (i.e., reactants, transition states, and products). The model was also validated by comparing the radial distribution functions calculated via the MLFF with available experimental and initio ones.
通过对六个独立系统(表 S1 和 S2)进行即时 AIMD 模拟,优化了用于构建 MLFF 的主要超参数。在近 个即时模拟步骤( )的基础上,使用主动学习算法对总共 7017 个关键第一原理点进行了采样。经过训练的 MLFF 模型分别由 4336、5053 和 2894 个基函数组成,用于 元素。MLFF 模型首先在未用于模型训练的外部数据集上进行了验证。该数据集包括 1800 个结构,涵盖不同的相位(即气相、水相和空气-水界面)和反应状态(即反应物、过渡态和产物)。该模型还通过比较通过 MLFF 计算出的径向分布函数与现有的实验 initio 函数进行了验证。
MLFF-Based Molecular Dynamics Simulation. The simulations were performed using VASP 6.3. The simulated system for the reactions at the air-water interface was a supercell that contains the reactants, molecules, and a vacuum layer (Figure S1). The aqueous phase system was a cubic cell with a side length that contained the reactants and molecules (Figure S2). The gaseous-phase system had the same size of the aqueous-phase system but only contains reactants (Figure S3). Free-energy profiles were calculated with several collective variables (CVs) using metadynamics sampling (detailed in the Supporting Information).
基于 MLFF 的分子动力学模拟。模拟使用 VASP 6.3 进行。 空气-水界面反应的模拟系统是一个 超级池,其中包含反应物、 分子和 真空层(图 S1)。水相系统是一个边长为 的立方晶胞,其中包含反应物和 分子(图 S2)。气相体系的大小与水相体系相同,但只包含反应物(图 S3)。利用元动力学采样(详见《辅助资料》)计算了多个集合变量(CV)的自由能曲线。
The simulation time is 1 ns for all considered systems in MLFFbased molecular dynamics simulation, and one initial configuration was adopted. An early study suggests that approximately nanosecond level simulation is long enough to produce the precise free-profile. Our test simulations indicated that the error caused by four initial configurations is very small ( , Figure S4). Gibbs-freeenergy change between separated reactants and the corresponding transition state ( ) was calculated by subtracting Gibbs-freeenergy at the transition state and the separated reactants. Detailed parameters for metadynamics samplings are listed in Table S3.
在基于 MLFF 的分子动力学模拟中,所有考虑的系统的模拟时间均为 1 毫微秒,并采用一种初始构型。早期研究表明,大约纳秒级的模拟时间足以产生精确的自由轮廓。 我们的试验模拟表明,四种初始构型造成的误差非常小( ,图 S4)。分离的反应物和相应过渡态之间的吉布斯自由能变化( )是通过减去过渡态和分离的反应物的吉布斯自由能计算出来的。表 S3 列出了元动力学取样的详细参数。
Quantum Chemical Calculations. Structural optimization and energy calculation were performed using Gaussian 16 and ORCA 4.0, respectively. For all the reactions involving and the organic molecules, geometry optimization and harmonic frequency calculation for the reactants, products, and transition states were carried out using the M06-2X-D3 /aug-cc-pVTZ methods. Intrinsic reaction coordinate calculation was performed to confirm the connection of the transition states with the reactants and products. Single-point energies were calculated at the aug-cc-pVTZ level (for " system) or DLPNO-CCSD aug-cc [for "HOCH " systems] based on the optimized geometries. The DLPNO-CCSD(T) aug-cc-pVTZ calculations were performed under the tightPNO condition. Zeropoint energies at the aug-cc-pVTZ level were adopted to correct the corresponding single-point energies.
量子化学计算。结构优化和能量计算分别使用高斯 16 和 ORCA 4.0 进行。 对于涉及 和有机分子的所有反应,使用 M06-2X-D3 /aug-cc-pVTZ 方法对反应物、产物和过渡态进行了几何优化和谐波频率计算。进行了本征反应坐标计算,以确认过渡态与反应物和生成物之间的联系。根据优化的几何结构,在 aug-cc-pVTZ 水平(针对 " 系统)或 DLPNO-CCSD aug-cc [针对 "HOCH " 系统] 下计算了单点能量。DLPNO-CCSD(T) aug-cc-pVTZ 计算是在紧PNO 条件下进行的。采用 aug-cc-pVTZ 水平的零点能量来修正相应的单点能量。
Figure 2. Adsorption of at air-water interfaces: (A) Gibbs-free-energy change ( ) profile of a molecule pulled into water as a function of distance between centers of and water slab [ density of water is indicated by the black curve and fitted via a hyperbolic tangent function; GDS indicates the Gibbs dividing surface, where equals bulk density of water.]; (B) probability distribution of nine typical configurations; and (C) their chemical structures at the water surface. (The white, gray, and red balls represent , and atoms, respectively. The dotted lines denote hydrogen bonds between and molecules.) (D) Definition of (E) probability density ( ) of at the water surface.
图 2.空气-水界面上的 吸附:(A) 吸入水中的 分子的吉布斯自由能变化 ( ) 曲线与 和水板中心间距离 的函数关系 [ 黑色曲线表示水的密度 ,并通过双曲正切函数拟合;GDS 表示吉布斯分界面,其中 等于水的体积密度 。];(B) 九种典型构型的概率分布;(C) 它们在水面上的化学结构。(白球、灰球和红球分别代表 原子和 原子。虚线表示 分子之间的氢键)。(D) 的定义 (E) 在水面的概率密度 ( )。

RESULTS AND DISCUSSION 结果与讨论

Performance of the Developed MLFF. To investigate the adsorption and transformation of at air-water interfaces, an MLFF model was trained based on a series of AIMD simulations with and without metadynamics samplings for gaseous, aqueous, and interfacial systems. An active-learning strategy that has succeeded in many aqueousphase, interface, and material systems was adopted to select key initio data for constructing the model. Figure 1 and Table S4 display performance of the trained MLFF model using SCAN-D4 level data as benchmarks, which is evaluated by mean absolute error (MAE), and root-mean- square error (RMSE) for training sets (i.e., data sets used for training the model) and external validation sets (i.e., data sets not used for training the model).
开发的 MLFF 的性能。为了研究 在空气-水界面上的吸附和转化,我们基于一系列有元动力学采样 和无元动力学采样 的 AIMD 模拟,对气体、水和界面系统进行了 MLFF 模型训练。我们采用了在许多水相、界面和材料系统中取得成功的主动学习策略来选择用于构建模型的关键 初始数据。 图 1 和表 S4 显示了以 SCAN-D4 级 数据为基准训练的 MLFF 模型的性能,该性能通过训练集(即用于训练模型的数据集)和外部验证集(即未用于训练模型的数据集)的平均绝对误差 (MAE) 和均方根误差 (RMSE) 进行评估。
In general, the trained MLFF model has good performance in energy (Figure 1A), force (Figure 1B), and stress (Figure 1C) predictions for the training (gray), and external validation (blue) sets. It can be seen that the MAE and RMSE values for the training and the external validation sets are immensely close (Table S4), indicating a high generalization ability of the model. The trained model also well reproduced experimental radial distribution functions of in in , and in , which were either determined
总体而言,经过训练的 MLFF 模型在训练集(灰色)和外部验证集(蓝色)的能量(图 1A)、力(图 1B)和应力(图 1C)预测方面表现良好。可以看出,训练集和外部验证集的 MAE 值和 RMSE 值非常接近(表 S4),表明模型具有很高的泛化能力。训练模型还很好地再现了 的实验径向分布函数。
Figure 3. Gaseous free-energy ( , kcal-mol ) landscapes for initial oxidation of by at (A) - group with (B) corresponding representative snapshots, and at (C) group and (D) corresponding representative snapshots. , and in panels A and C represent approximate locations of the observed separated reactants, reactants, transition states, and products, respectively. The contour lines in panels and are spaced apart. , and in panels and stand for coordinate numbers of the atom in (excluding the original of ), the atoms in , and the atom in , respectively. The white, gray, and red balls in panels and stand for , and atoms, respectively.]
图 3.在 (A) - 组和 (B) 相应的代表性快照,以及在 (C) 组和 (D) 相应的代表性快照中, 初始氧化的气体自由能 ( , kcal-mol ) 分布图。面板 A 和 C 中的 分别代表观察到的分离反应物、反应物、过渡态和产物的大致位置。面板 中的等高线间距为 。面板 中的 分别代表 原子(不包括 中的原始 原子)、 原子和 原子的 坐标。面板 中的白球、灰球和红球分别代表 原子]。
experimentally (Figure 1D) or calculated via the AIMD simulations (Figure 1E,F). Hence, the MLFF model can be employed to simulate the behavior of at air-water interfaces.
(图 1D)或通过 AIMD 模拟计算得出的结果(图 1E、F)。因此,MLFF 模型可用于模拟 在空气-水界面上的行为。
Enrichment of on Water Surfaces. Gibbsfree-energy changes for transferring an molecule from the gaseous phase to the interior region of water were computed using the trained MLFF, as shown in Figure . Based on the between the gaseous and aqueous phases, Henry's law constant for was calculated to be (see details in the Supporting Information) and located in a previously determined range to . Due to the hydrophobic nature of the group and hydrogen bonding of the group in the molecule, the minimum of was observed at the interface rather than in the interior region of water or the gaseous phase. The results indicate a notable "surface preference" of , like its dehydration form .
水表面 的富集。如图 所示,使用训练有素的 MLFF 计算了将 分子从气相转移到水内部区域的吉布斯自由能变化 。根据气相和水相之间的 ,计算出 的亨利定律常数 (详见《辅助信息》),并位于先前确定的