Machine Learning: Deep Learning Spectroscopy: Neural Networks for Molecular Excitation Spectra (Adv. Sci. 9/2019)

Ghosh, Kunal; Stuke, Annika; Todorović, Milica; Jørgensen, Peter Bjørn; Schmidt, Mikkel N.; Vehtari, Aki; Rinke, Patrick

doi:10.1002/advs.201970053

Cited by 15 publications

(21 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Apart from being a useful reference for 2D materials research, our library can be used to train machine learning algorithms for materials science problems . Similarly to recent work on prediction of linear optical spectra for molecules, our database may enable prediction of NLO spectra directly from the atomic structure and, in turn, autonomous ( in situ ) characterization of materials …”

Section: Discussionmentioning

confidence: 99%

Two-Dimensional Materials with Giant Optical Nonlinearities near the Theoretical Upper Limit

2021

View full text Add to dashboard Cite

Nonlinear optical (NLO) phenomena such as harmonic generation and Kerr and Pockels effects are of great technological importance for lasers, frequency converters, modulators, switches, etc. Recently, two-dimensional (2D) materials have drawn significant attention due to their strong and peculiar NLO properties. Here, we describe an efficient first-principles workflow for calculating the quadratic optical response and apply it to 375 non-centrosymmetric semiconductor monolayers from the Computational 2D Materials Database (C2DB). Sorting the nonresonant nonlinearities with respect to bandgap E g reveals an upper limit proportional to E g –4, which is neatly explained by two distinct generic models. We identify multiple promising candidates with giant nonlinearities and bandgaps ranging from 0.4 to 5 eV, some of which approach the theoretical upper limit and greatly outperform known materials. Our comprehensive library of ab initio NLO spectra for all 375 monolayers is freely available via the C2DB Web site.

show abstract

Section: Discussionmentioning

confidence: 99%

Two-Dimensional Materials with Giant Optical Nonlinearities near the Theoretical Upper Limit

2021

View full text Add to dashboard Cite

show abstract

“…Apart from being a useful reference for 2D materials research, our library can be used to train machine learning algorithms for materials science problems [73]. Similarly to recent work on prediction of linear optical spectra for molecules [74], our database may enable prediction of NLO spectra directly from the atomic structure and, in turn, autonomous (in situ) characterization of materials [75].…”

Section: Discussionmentioning

confidence: 99%

Two-dimensional Materials with Giant Optical Nonlinearities Near the Theoretical Upper Limit

Taghizadeh,

Thygesen,

Pedersen

2020

Preprint

View full text Add to dashboard Cite

Nonlinear optical (NLO) phenomena such as harmonic generation, Kerr, and Pockels effects are of great technological importance for lasers, frequency converters, modulators, switches, etc. Recently, two-dimensional (2D) materials have drawn significant attention due to their strong and unique NLO properties. Here, we describe an efficient first-principles workflow for calculating the quadratic optical response and apply it to 375 non-centrosymmetric semiconductor monolayers from the Computational 2D Materials Database (C2DB). Sorting the non-resonant nonlinearities with respect to bandgap Eg reveals an upper limit proportional to E −4 g , which is neatly explained by two distinct generic models. We identify multiple promising candidates with giant nonlinearities and bandgaps ranging from 0.4 to 5 eV, some of which approach the theoretical upper limit and greatly outperform known materials. Our comprehensive library of ab initio NLO spectra for all 375 monolayers is freely available via the C2DB website. We expect this work to pave the way for highly efficient and compact opto-electronic devices based on 2D materials.

show abstract

“…The original dataset contained a few very small molecules, such as CH 4 , for which some of the lowest-energy excitations lie far below −30 eV (in the range around −200 eV). 10 We consider 12 such small molecules as outliers and clean the dataset by removing these molecules, leaving 132,519 molecules for our evaluation.…”

Section: ■ Methodsmentioning

confidence: 99%

“…8 Early efforts for the ML-based prediction of molecular spectra have been presented elsewhere, 9 whereby TDDFTcalculated UV spectra were used to train machine learning models. A cornerstone work 10 showed that the usage of deep tensor neural networks (DTNN) 11 leads to new state-of-theart results for the prediction of spectra of the frequently used QM9 dataset of organic molecules. DTNNs are graph neural networks (GNNs) that represent the molecules under study with matrices representing charges and distances.…”

Section: ■ Introductionmentioning

confidence: 99%

Graph Neural Networks for Learning Molecular Excitation Spectra

Singh

Münchmeyer

Weber

et al. 2022

J. Chem. Theory Comput.

View full text Add to dashboard Cite

Machine learning (ML) approaches have demonstrated the ability to predict molecular spectra at a fraction of the computational cost of traditional theoretical chemistry methods while maintaining high accuracy. Graph neural networks (GNNs) are particularly promising in this regard, but different types of GNNs have not yet been systematically compared. In this work, we benchmark and analyze five different GNNs for the prediction of excitation spectra from the QM9 dataset of organic molecules. We compare the GNN performance in the obvious runtime measurements, prediction accuracy, and analysis of outliers in the test set. Moreover, through TMAP clustering and statistical analysis, we are able to highlight clear hotspots of high prediction errors as well as optimal spectra prediction for molecules with certain functional groups. This in-depth benchmarking and subsequent analysis protocol lays down a recipe for comparing different ML methods and evaluating dataset quality.

show abstract

Machine Learning: Deep Learning Spectroscopy: Neural Networks for Molecular Excitation Spectra (Adv. Sci. 9/2019)

Cited by 15 publications

References 0 publications

Two-Dimensional Materials with Giant Optical Nonlinearities near the Theoretical Upper Limit

Two-Dimensional Materials with Giant Optical Nonlinearities near the Theoretical Upper Limit

Two-dimensional Materials with Giant Optical Nonlinearities Near the Theoretical Upper Limit

Graph Neural Networks for Learning Molecular Excitation Spectra

Contact Info

Product

Resources

About