Data-driven Approaches for Predicting Excited-state Properties in Condensed Phase Molecular Systems Open Access

Ren, Fangning (Spring 2026)

Permanent URL: https://etd.library.emory.edu/concern/etds/6682x558t?locale=en
Published

Abstract

Predicting excited-state properties of molecular systems in condensed phase environments remains a challenge in theoretical chemistry, where the interplay between environmental effects and conformational diversity imposes severe computational bottlenecks. This dissertation develops and applies data-driven approaches that address three key problems in this field.

First, we resolve the long-lasting question of which optimal tuning scheme for range-separated hybrid (RSH) density functionals yield the most accurate solution-phase excitation energies. We evaluated multiple gamma-tuning protocols—gas-phase gamma tuning (GPγT), partial vertical gamma tuning (PVγT), strict vertical gamma tuning (SVγT), and optimally tuned screened RSH functional with polarizable continuum model (SRSH-PCM)—against experimental UV/Vis absorption data for 937 solvated molecules spanning 14 solvents. The analysis reveals that SVγT achieves a mean absolute error of 0.35 eV, outperforming other protocols. Furthermore, we found that the smaller γ values from SVγT captured the expected 1/εr asymptotic behavior in the solution phase, resulting in accurate prediction of solution-phase CT excitations.

Second, we apply the extended Frenkel exciton model with charge-transfer (CT) states to investigate the fluorescence properties of asphaltene aggregates. Through large-scale calculations on over 500 asphaltene dimer configurations encompassing 133 monomers, we demonstrate that dimer fluorescence never exceeds both of its monomers, with further redshifts in heterodimers are primarily caused by LE-CT coupling. These findings provide the first systematic, data-driven understanding of PAH aggregate photo physics across a chemically diverse compositional space.

Third, we introduce a size-transferable machine learning (ML) method based on Frenkel exciton model that predicts excited-state properties for molecular assemblies of arbitrary size using models trained only on dimer Hamiltonians. We found that the exciton model Hamiltonian of large aggregates can be evaluated by conducting QC calculation on all dimer pairs in the assemblies, allowing an ML model trained on dimers to reconstruct Hamiltonians for aggregates of any size. We also proposed a new method to address the phase-correction problem by introducing coupling terms’ approximations. Our model accurately predicted the excitation energies of trimer and tetramer of perylene and tetracene and estimated S1 oscillator strengths of perylene aggregates. Leveraging our ML model, the optical gaps of nanosized perylene aggregates with up to 50 monomers are analyzed, qualitatively revealing the role of different types of couplings on their size dependency.

Together, these studies demonstrate that data-driven methodologies—spanning high-throughput benchmarking, statistical analysis of large conformational ensembles, and machine learning—can overcome fundamental barriers in condensed-phase excited-state modeling that are intractable with conventional quantum chemistry approaches alone.

Table of Contents

1   Chapter 1 : Introduction

1   1.1 Electronic Excited-states and Theoretical Modeling

2   1.2 Challenges Facing Quantum Chemistry Methods

4   1.3 Data-Driven Approaches and the Overview

6   References

10  Chapter 2 : Theoretical Background

10  2.1 Density Functional Theory and Time-Dependent Extensions

10  2.1.1 The Electronic Structure Problem and Hartree-Fock Theory

12  2.1.2 Density Functional Theory

15  2.1.3 Time-Dependent Density Functional Theory

20  2.1.4 Range-Separated Hybrid Functionals and Optimal Tuning

23  2.2 Solvation Models

23  2.2.1 The Polarizable Continuum Model

25  2.2.2 Explicit Solvation: Molecular Dynamics and QM/MM

26  2.3 The Frenkel Exciton Model

26  2.3.1 The Standard Frenkel Exciton Model

27  2.3.2 The Extended Frenkel Exciton Model with Charge-Transfer States

31  References

36  Chapter 3 : Data-Driven Evaluation of Optimal Tuning Schemes for Range-Separated Hybrid Functionals in Solution

36  3.1 Introduction

43  3.2 Method

43  3.2.1 Dataset curation

45  3.2.2 DFT calculations

48  3.2.3 γ-tuning procedure

49  3.2.4 Asymptote behavior evaluation

51  3.3. Results

51  3.3.1 The optimum γ-value distribution

58  3.3.2 Assessing the γ-tuning procedure

69  3.3.3 Impact of solutes and solvents

76  3.3.5 Impact of γ-tuning on delocalization error.

78  3.3.4 Reasons for the better performance of PVγT and SVγT

85  3.4. Conclusion

88  3.5 Appendix

88  3.5.1 Proof of the impact of PCM on the HOMO energy and IP for the N+1 anionic state

91  3.5.2 Assessing the impact of solvent polarity on optimal γ under SVγT.

92  3.6 References

102 Chapter 4 Data-Driven Insights into Asphaltene Aggregate Fluorescence Using the Extended Frenkel Exciton Model

102 4.1 Introduction

105 4.2 Theory

110 4.3 Methods

110 4.3.1 Preparation of asphaltene monomers

111 4.3.2 Generation of initial asphaltene dimer structures

114 4.3.3 DFT & TDDFT calculations

118 4.3.4 Exciton model and electron excitation analysis

119 4.4 Results

119 4.4.1 Monomer properties

121 4.4.2 The C22-PC113 Dimer

126 4.4.3 The C05-C22 Dimer

130 4.4.4 Fluorescence energy of the C22-X and PC113-X series

135 4.4.5 Oscillator strength of the C22-X series

138 4.4.6 Fluorescence and oscillator strength of 465 coal asphaltene dimers.

144 4.5 Conclusion

147 4.6 Appendix

152 4.7 References

161 Chapter 5 Size-Transferable Prediction of Excited-state Properties with Machine-Learned Exciton Models

161 5.1 Introduction

165 5.2 The Extended Frenkel Hamiltonian for Large Assemblies

166 5.2.1 Hamiltonian Structure of a N-mer (an assembly has N monomers)

168 5.2.2 Pairwise Decomposability of Matrix Elements

170 5.2.3 Approximating the Type 1 CT-CT coupling

173 5.2.4 Investigation of the proportionality factor for Type 1 CT-CT couplings

176 5.3 Physical Approximations for the Hamiltonian Matrix Elements

176 5.3.1 Motivation: a Δ-ML Strategy for the Exciton Hamiltonian

177 5.3.2 Reference Wavefunction Alignment

178 5.3.3 Approximations for Each Hamiltonian Term

180 5.3.4 Decomposition of Approximations into Atomic Contributions

181 5.4 Training Dataset Construction and Reference Calculations

181 5.4.1 Molecular Dynamics Sampling for Dimer Extraction

182 5.4.2 SEP5A: Constructing the Separated-Dimer Subset via Approximations

185 5.4.3 QC Calculations

186 5.4.4 Data Augmentation via Belonging-Label Swap

186 5.5 Solving the Phase Problem for Large Aggregates

189 5.6 Machine Learning Model Architecture and Prediction Workflow

189 5.6.1 Modified TorchANI Architecture Overview

193 5.6.2 The LE network.

193 5.6.3 The CT network.

194 5.6.4 The coupling networks.

194 5.6.5 Training Procedure and Hyperparameters

195 5.7 Results

195 5.7.1 Dimer Hamiltonian Accuracy

197 5.7.2 Excited-state Energy Prediction for Trimers and Tetramers

199 5.7.3 Oscillator Strength Prediction for Trimers and Tetramers

203 5.8 Application: Optical Gap Size Dependency of Perylene Nanoaggregates

204 5.8.1 Optical Gap vs. Aggregate Size

207 5.8.2 Decomposition of Coupling Contributions

210 5.9 Discussion, Limitations, and Future Directions

211 5.10 Conclusion

About this Dissertation

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files