TY - CHAP
T1 - Design of organic materials with tailored optical properties
T2 - Predicting quantum-chemical polarizabilities and derived quantities
AU - Vishwakarma, Gaurav
AU - Sonpal, Aditya
AU - Pradhan, Aatish
AU - Haghighatlari, Mojtaba
AU - Afzal, Mohammad Atif Faiz
AU - Hachmann, Johannes
N1 - Publisher Copyright:
© 2023 Elsevier Inc. All rights reserved.
PY - 2022/1/1
Y1 - 2022/1/1
N2 - In this chapter, we present a rational design protocol for the discovery and design of new materials and chemistry. With a focus on optical materials with a high refractive index, we describe four components of this protocol that we believe are crucial for its success—molecular library generation, virtual high-throughput screening, database management, and data mining for subsequent analyses. Through a handful of case studies, we demonstrate the effectiveness of our group's software ecosystem, which is aimed at realizing this rational design protocol. In the process, we explore several state-of-the-art machine learning (ML) methods for accelerating the discovery of new materials, such as a multitask, physics-infused deep learning model, and a transfer learning model for tackling the inaccuracies involved with smaller datasets. We show that these models outperform other trivial machine learning models by a significant margin and that incorporating known physics into data-derived models provides valuable guardrails.
AB - In this chapter, we present a rational design protocol for the discovery and design of new materials and chemistry. With a focus on optical materials with a high refractive index, we describe four components of this protocol that we believe are crucial for its success—molecular library generation, virtual high-throughput screening, database management, and data mining for subsequent analyses. Through a handful of case studies, we demonstrate the effectiveness of our group's software ecosystem, which is aimed at realizing this rational design protocol. In the process, we explore several state-of-the-art machine learning (ML) methods for accelerating the discovery of new materials, such as a multitask, physics-infused deep learning model, and a transfer learning model for tackling the inaccuracies involved with smaller datasets. We show that these models outperform other trivial machine learning models by a significant margin and that incorporating known physics into data-derived models provides valuable guardrails.
KW - Machine learning
KW - Physics-infused
KW - Polarizability
KW - Rational design
KW - Refractive index
KW - Transfer learning
UR - http://www.scopus.com/inward/record.url?scp=85142726267&partnerID=8YFLogxK
U2 - 10.1016/B978-0-323-90049-2.00028-7
DO - 10.1016/B978-0-323-90049-2.00028-7
M3 - Chapter
AN - SCOPUS:85142726267
SN - 9780323886048
SP - 653
EP - 674
BT - Quantum Chemistry in the Age of Machine Learning
PB - Elsevier
ER -