Machine Learning Assisted Understanding and Discovery of CO₂ Reduction Reaction Electrocatalyst

Abstract

Electrochemical CO 2 reduction reaction (CO 2 RR) is an important process which is a potential way to recycle excessive CO 2 in the atmosphere. Although the electrocatalyst is the key toward efficient CO 2 RR, the progress of discovering effective catalysts is lagging with current methods. Because of the cost and time efficiency of the modern machine learning (ML) algorithm, an increasing number of researchers have applied ML to accelerate the screening of suitable catalysts and to deepen our understanding in the mechanism. Hence, we reviewed recent applications of ML in the research of CO 2 RR by the types of electrocatalyst. An introduction on the general methodology and a discussion on the pros and cons for such applications are included.

Electrochemical reactions catalyzed by carbon dots from computational investigations: functional groups, dopants, and defects

“…The application of the machine learning interatomic potential signicantly reduces the computational cost and has potential for simulation in real time. [209][210][211] Hajibabaei and Kim applied sparse Gaussian process potentials to explore the conductivity of the electrolyte. 203,212 The concept of sparse Gaussian process potentials (SGPR) exploits the similarity of the chemical environment, which signicantly decreases the O(n 3 ) scaling of conventional Gaussian process potentials.…”

Computational Chemistry as Applied in Environmental Research: Opportunities and Challenges

“…Different subsets of machine learning methodologies are classified into supervised, unsupervised, and reinforced learning. − In supervised learning methods, such as linear and logistic regression, decision trees, support vector machines, Naïve Bayes, k-nearest neighbors, neural networks, and random forest techniques, the algorithms are built using labeled data where each input data are associated with a target value . K-means clustering, principal component analysis, and generative adversarial networks are examples of unsupervised learning techniques, which, in contrast, do not require labeled data to learn patterns or extract relationships within the data (i.e., the data set is composed only of input data) . Reinforcement learning ML methodologies acquire knowledge by interacting with an environment in which the agent (learner) interacts to receive cumulative reward (or penalty) signals.…”

“…311 K-means clustering, principal component analysis, and generative adversarial networks are examples of unsupervised learning techniques, which, in contrast, do not require labeled data to learn patterns or extract relationships within the data (i.e., the data set is composed only of input data). 310 Reinforcement learning ML methodologies acquire knowledge by interacting with an environment in which the agent (learner) interacts to receive cumulative reward (or penalty) signals. In this scenario, the training data do not provide specific correct output for inputs but indicate whether an action is correct (reward) or not (penalty), making the ultimate goal of the model to maximize the expected rewards over time.…”

Accessing the usefulness of atomic adsorption configurations in predicting the adsorption properties of molecules with machine learning

“…Machine learning (ML) emerged recently as a tool to accelerate the study of interfaces, especially the adsorption of small molecules on surfaces. Studies have successfully predicted properties of the adsorbate/substrate system such as adsorption energy, [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15] reaction barrier, 16 adsorption height, 17 and buckling of the surface. 17 The main application in many of these studies has been heterogenous catalysis where one can employ adsorption energy or another adsorption property as a chemical descriptor, [18][19][20][21] or an indicator of catalytic activity.…”