Permanganate Reaction Kinetics and Mechanisms and Machine Learning Application in Oxidative Water Treatment PDF Download

Author: Shifa Zhong
Publisher:
ISBN:
Category : Chemical kinetics
Languages : en
Pages : 256

Get Book

Book Description
Permanganate (MnO4-) plays an important role in water treatment as a strong oxidant. Two additives, i.e., bisulfite (HSO3-) and ligands, have been found to significantly accelerate its oxidation rates toward organic contaminants, but the specific mechanisms remain largely unknown or controversial. Reaction rate constants of contaminants toward various oxidants or reductants are an important parameter for optimizing water/wastewater treatment; however, experimentally measuring rate constants for thousands of contaminants is time-consuming and labor-intensive. In comparison, developing quantitative structure activity relationship (QSAR) models for estimating their rate constants is an efficient approach with satisfactory accuracy. The presence of bisulfite can make the oxidation of organic compounds by MnO4- complete in milliseconds. Previous studies concluded that uncomplexed Mn(III) was responsible for this millisecond reactivity. However, we revealed that this ultrafast reactivity was only observed in the presence of O2. We also found that HSO3- and O2 were rapidly consumed when mixing HSO3- with MnO4- in the presence of O2. This was because reactive Mn intermediates, mainly Mn(III) species, were generated in situ from the reaction of HSO3- and MnO4-, which then acted as a catalyst for the reaction of HSO3- and O2. In the presence of organic compounds, this catalytic effect was weakened because the reactive Mn intermediates were consumed by reacting with the organic compounds. However, without O2 these reactive Mn intermediates cannot oxidize the organic compounds. Hence, we concluded that only the Mn(III) with this catalytic role can oxidize organic compounds in milliseconds. This work unveiled the important role of O2 in the HSO3-/MnO4- system, which is important for its real applications. Ligands, such as pyrophosphate (PP), nitrilotriacetate (NTA), and ethylenediaminetetraacetic acid (EDTA), are known to increase the oxidation reactivity of phenolic compounds by MnO4- by several times. The traditional explanation for this acceleration effect is that ligands can complex with the Mn(III) intermediates being generated from the reaction of MnO4- and phenolic compounds to form Mn(III)-ligand complexes, and these complexes then oxidize phenolic compounds much faster than MnO4- can. Here, we observed that Mn(III)-ligand complexes formed during the reaction but were not further consumed. We then used pentachlorophenol (PCP) as a probe because it can be oxidized by MnO4- but not by Mn(III)-ligand complexes. In the presence of these complexes, the oxidation rate of PCP by MnO4- was accelerated. Hence, we proposed a new reaction mechanism in which Mn(III)-ligand complexes also act as a catalyst for the reaction of MnO4- and phenolic compounds. This work gave another explanation to the effect of ligands on MnO4-, which will substantially benefit the application of MnO4-/ligand systems in water/wastewater treatment. Not only for MnO4- but also for other common oxidants, e.g., O3, HO• and SO4•- radicals, the reaction rate constant is an important parameter for optimizing the treatment process, such as determining the dosage of an oxidant or the treatment time. However, it is time-consuming and labor-intensive to experimentally measure the rate constants of thousands of organic compounds. Toward this end, quantitative structure−activity relationships (QSARs) have been widely employed to correlate chemical structures of compounds with their reactivity. Well-calibrated QSARs can help predict the rate constants of a large number of organic contaminants based on their chemical structures and have played important roles in many environmental applications, such as estimating the rate constants or chemical toxicity. We here introduced molecular fingerprints (MF) to represent various organic contaminants and combined them with machine learning algorithms to develop QSAR models. We compared the predictive performance of MF-based QSAR models with that of MD-based ones, and found that their predictive performance was comparable, thus demonstrating the effectiveness of MF-based QSAR models. Due to the "black box" nature of machine learning algorithms in general, we then interpreted the MF-based machine learning QSAR models by the Shapley Additive Explanation (SHAP) method. Results showed that MF-based machine learning QSAR models made prediction on the rate constants based on the correct understanding of how the atom groups affect the rate constants, such as the effect of electron-donating and electron-withdrawing groups, thus demonstrating that the MF-base machine learning models were trustful. Apart from the molecular fingerprints, we also employed 2D molecular images to represent organic compounds and combined them with a convolutional neural network (CNN) to develop QSAR models. When developing CNN-based QSARs, we applied transfer learning and data augmentation to further enhance the predictive performance and robustness of the model. We also interpreted the obtained molecular image-CNN model by the Gradient-weighted Class Activation Mapping (Grad-CAM) technique, and the results showed that our model makes predictions by choosing correct features in the molecular images. Overall, this work introduced two new representations for organic contaminants, which have not been reported in the environmental field before, and the interpretations for the QSAR models offered some much-needed theoretical support for trusting these models. Overall, the new findings on HSO3-/MnO4- further elucidate why HSO3-/MnO4- is so reactive, especially regarding the key role of O2. In real applications, supplying enough O2, such as bubbling with air, is necessary for this system to achieve high efficiency. The new findings on MnO4-/ligand illustrate how ligands accelerate the oxidation of phenolic compounds by MnO4- and imply that Mn(III)-ligand complexes may be released into water and continue to facilitate the oxidation of compounds that cannot be oxidized by Mn(III)-ligand alone. In real applications, we should pay more attention to compounds that can be oxidized by MnO4- rather than by Mn(III)-ligand, because Mn(III)-ligand mainly acts as a catalyst. For QSAR model development, we introduced two new representations for contaminants, namely, molecular fingerprints and molecular images, to combine with machine learning to develop QSAR models for predicting the rate constants of contaminants toward HO• radicals. Reactivity of contaminants in AOPs can be more easily estimated with these QSAR models. Transfer learning, data augmentation and model interpretation are three important concepts that can be applied to other QSAR models, such as predicting plant uptake or toxicity. Any QSAR models that involve chemicals can benefit from our study, that is, representing the chemicals by molecular fingerprints or molecular images, applying transfer learning and data augmentation, and interpreting the QSAR models.