Kernel Methods for Quantum Chemistry

Pronobis, Wiktor; Müller, Klaus Robert

doi:10.1007/978-3-030-40245-7_3

Cited by 4 publications

(5 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The smooth overlap of Article atomic positions (SOAP) 123,124 is another structure-sensitive descriptor considered in this work, which can be used to compute the similarity between a pair of local atomic environments-and, by extension, a pair of structures-by representing the atoms as Gaussians (i.e., ''smoothed positions'') and comparing the spatial overlap in the resulting atomic density fields (Figure S2 and Equations S5-S9). In all of the aforementioned examples, these features are used to develop a kernel ridge regression 125 (KRR) model (Equations S1-S3). Motivated by prior work on inorganic solids, we also investigated the use of a crystal graph convolutional neural network (CGCNN), 68 wherein an approximate crystal graph is generated for each MOF, with each node in the graph representing an atom and each edge representing the bonds that connect the atoms (Figure 3F).…”

Section: Ll Open Accessmentioning

confidence: 99%

Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery

Rosen

Iyer

Ray

et al. 2021

Matter

251

259

View full text Add to dashboard Cite

Using a new database of electronic structure properties derived from highthroughput density functional theory calculations for thousands of metal-organic frameworks (MOFs), we benchmark a variety of machine learning models to accurately and rapidly predict MOF band gaps. Unsupervised dimensionality reduction techniques are also used to map the MOF feature space and identify otherwise subtle structure-property relationships. We anticipate that machine learning models derived from this new database will accelerate the discovery of promising MOFs with targeted quantum-chemical properties.

show abstract

Section: Ll Open Accessmentioning

confidence: 99%

Machine learning the quantum-chemical properties of metal–organic frameworks for accelerated materials discovery

Rosen

Iyer

Ray

et al. 2021

Matter

251

259

View full text Add to dashboard Cite

show abstract

“…atomic positions (SOAP) 123,124 is another structure-sensitive descriptor considered in this work, which can be used to compute the similarity between a pair of local atomic environments-and, by extension, a pair of structures-by representing the atoms as Gaussians (i.e., ''smoothed positions'') and comparing the spatial overlap in the resulting atomic density fields (Figure S2 and Equations S5-S9). In all of the aforementioned examples, these features are used to develop a kernel ridge regression 125 (KRR) model (Equations S1-S3). Motivated by prior work on inorganic solids, we also investigated the use of a crystal graph convolutional neural network (CGCNN), 68 wherein an approximate crystal graph is generated for each MOF, with each node in the graph representing an atom and each edge representing the bonds that connect the atoms (Figure 3F).…”

Section: Ll Open Accessmentioning

confidence: 99%

“…The overlap of these density fields, integrated over all three-dimensional rotations (Equations S6-S7), are compared between structures to generate a kernel matrix describing the similarity between every pair of MOF structures (Figure S2 and Equations S8-S9). In all of the aforementioned examples, these features are used to develop a kernel ridge regression 114 (KRR) model (Equations S1-S3). Motivated by prior work on inorganic solids, we also investigated the use of a crystal graph convolutional neural network (CGCNN), 65 wherein an approximate crystal graph is generated for each MOF, with each node in the graph representing an atom and each edge representing the bonds that connect the atoms (Figure 3f).…”

Section: Machine Learning Models For Band Gap Predictionmentioning

confidence: 99%

Machine Learning the Quantum-Chemical Properties of Metal–Organic Frameworks for Accelerated Materials Discovery with a New Electronic Structure Database

Rosen¹,

Iyer²,

Ray³

et al. 2020

Preprint

View full text Add to dashboard Cite

<p>Metal–organic frameworks (MOFs) are a widely investigated class of crystalline solids with tunable structures that make it possible to impart specific chemical functionality tailored for a given application. However, the enormous number of possible MOFs that can be synthesized makes it difficult to determine which materials would be the most promising candidates, especially for applications governed by electronic structure properties that are often computationally demanding to simulate and time-consuming to probe experimentally. Here, we have developed the first publicly available quantum-chemical database for MOFs (the “QMOF database”), which consists of properties derived from density functional theory (DFT) for over 14,000 experimentally synthesized MOFs. Throughout this study, we demonstrate how this new database can be used to identify MOFs with targeted electronic structure properties. As a proof-of-concept, we use the QMOF database to evaluate the performance of several machine learning models for the prediction of DFT-computed band gaps and find that crystal graph convolutional neural networks are capable of achieving superior predictive performance, making it possible to circumvent computationally expensive quantum-chemical calculations. We also show how unsupervised learning methods can aid the discovery of otherwise subtle structure–property relationships using the computational findings in this work. We conclude by highlighting several MOFs with low band gaps, a challenging task given the electronically insulating nature of most MOF structures. The data and predictive models generated in this work, as well as the database of MOF structures, should be highly useful to other researchers interested in the predictive design and discovery of MOFs for the many applications dictated by quantum-chemical phenomena.<br></p>

show abstract

“…Over the past several decades several ML-based methods have been used to represent continuous PESs. 3,15–17 While a number of those are briefly mentioned below, the focus of the present work is on NN-based approaches. Kernel-based methods provide an efficient solution to highly non-linear optimization problems 17 by finding a representation of the problem which encodes the distribution of the data in a complete, unique and efficient way.…”

Section: Introductionmentioning

confidence: 99%

“…3,15–17 While a number of those are briefly mentioned below, the focus of the present work is on NN-based approaches. Kernel-based methods provide an efficient solution to highly non-linear optimization problems 17 by finding a representation of the problem which encodes the distribution of the data in a complete, unique and efficient way. 18 There is a large number of possible representations of chemical space that can be used in kernel methods.…”

Section: Introductionmentioning

confidence: 99%