Lin Shen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Lin Shen has a total of 142 co-authors. Here are the top co-authors by number of shared publications:
- Wei-Hai Fang
College of Chemistry, Key Laboratory of Theoretical & Computational Photochemistry of Ministry of Education, Beijing Normal University
26 shared publications - Diandong Tang
Department of Chemistry, University of Washington, Seattle, Washington 98195, United States
10 shared publications Wei-Hai Fang
10 shared publications- Bin-Bin Xie
Zhejiang Normal University
7 shared publications - Weitao Yang
Department of Chemistry, Duke University
7 shared publications - Xuebo Chen
College of Chemistry and Molecular Sciences, Henan University, Kaifeng, 475004, China
7 shared publications
Latest Publications
- 23 September 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 17 (40) , 11606-11613
Probing Unusual Intrinsic Photophysical Properties of Thiocoumarin from XMS-CASPT2 Calculations
Aromatic thioketones have long served as model molecules for studying the fundamental principles of molecular photochemistry because their intrinsic photophysical properties differ markedly from those of aromatic ketones and many other aromatic compounds containing various organic functional groups. Thiocoumarin (TC), however, is an exceptional aromatic thioketone whose excited-state dynamics are distinctly different from those of common aromatic thioketones. Although many experimental and theoretical investigations have been devoted to TC and its derivatives, the intrinsic origin of their unusual photophysical properties remains unclear. In this work, high-level XMS-CASPT2 calculations were performed for the five lowest-lying electronic states, S0, T1, S1, T2, and S2, of TC and benzopyranthione (BPT), with BPT used as a representative common aromatic thioketone for comparison. The calculated results show that the small difference between the molecular structures of BPT and TC leads to substantial changes in the excited-state structures, electronic characters, and energy-level ordering in going from BPT to TC. These changes are identified as the intrinsic origin of the anomalous photophysical properties of TC. Meanwhile, the XMS-CASPT2 calculations reveal that the T1 state of TC can be populated through the two ultrafast S2 → S1 → T1 and S2 → T2 → T1 processes, whereas the subsequent T1 → S0 radiative and nonradiative processes are predicted to be inefficient. The theoretical predictions are in excellent agreement with the experimental findings that the long-lived T1 state of TC is formed with near-unity quantum yield. The T1-state features make TC and its derivatives metal-atom-free photosensitizers applicable in the field of photodynamic therapy.
- 23 September 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 17 (40) , 11606-11613
Probing Unusual Intrinsic Photophysical Properties of Thiocoumarin from XMS-CASPT2 Calculations
Aromatic thioketones have long served as model molecules for studying the fundamental principles of molecular photochemistry because their intrinsic photophysical properties differ markedly from those of aromatic ketones and many other aromatic compounds containing various organic functional groups. Thiocoumarin (TC), however, is an exceptional aromatic thioketone whose excited-state dynamics are distinctly different from those of common aromatic thioketones. Although many experimental and theoretical investigations have been devoted to TC and its derivatives, the intrinsic origin of their unusual photophysical properties remains unclear. In this work, high-level XMS-CASPT2 calculations were performed for the five lowest-lying electronic states, S0, T1, S1, T2, and S2, of TC and benzopyranthione (BPT), with BPT used as a representative common aromatic thioketone for comparison. The calculated results show that the small difference between the molecular structures of BPT and TC leads to substantial changes in the excited-state structures, electronic characters, and energy-level ordering in going from BPT to TC. These changes are identified as the intrinsic origin of the anomalous photophysical properties of TC. Meanwhile, the XMS-CASPT2 calculations reveal that the T1 state of TC can be populated through the two ultrafast S2 → S1 → T1 and S2 → T2 → T1 processes, whereas the subsequent T1 → S0 radiative and nonradiative processes are predicted to be inefficient. The theoretical predictions are in excellent agreement with the experimental findings that the long-lived T1 state of TC is formed with near-unity quantum yield. The T1-state features make TC and its derivatives metal-atom-free photosensitizers applicable in the field of photodynamic therapy.
- Open Access
- 22 July 2026
- Published by Scilight Press Pty Ltd in Materials Matter
- Vol. 1 (1) , 5
Machine Learning-Assisted Optimization and Application of Carbon-Based Emitters
Machine learning (ML) is progressively being integrated into materials science, exhibiting great potential for optimizing chemical synthesis and structural regulation, thereby accelerating intelligent design and efficient exploration of novel materials. Carbon-based emitters (CBEs), as emerging functional materials, have attracted considerable attention due to their tunable photoluminescence properties, abundant precursor sources, and structural versatility. To date, extensive research on CBEs has generated a substantial data foundation, laying the groundwork for ML driven structural screening and property prediction. Although the application of ML in CBEs remains in its early stages compared to that of inorganic semiconductors and organic-inorganic hybrid perovskites, its potential to accelerate material screening and reveal structure-property relationships is becoming increasingly evident. Given the transformative role ML has played in other functional materials, it is expected to drive a paradigm shift in CBEs research from conventional trial-and-error approaches to data-driven, intelligence-guided design, substantially accelerating material discovery and expanding their functional applications. Therefore, this review systematically summarizes recent advances in ML applications in CBEs, focusing on the general ML workflow, property prediction and structural design strategies for organic small-molecules and carbon dots (CDs), and their applications in organic light-emitting diodes (OLEDs), quantum dot light-emitting diodes (QLEDs), information encryption, biomedicine, and sensing. Finally, this review discusses the key challenges currently facing the field and offers perspectives on future directions for ML-driven CBEs research, aiming to provide guidance for the rational design and efficient development of such materials.
- 22 July 2026
- Published by Scilight Press Pty Ltd in Materials Matter
- Vol. 1 (1) , 5
Open AccessMachine Learning-Assisted Optimization and Application of Carbon-Based Emitters
Machine learning (ML) is progressively being integrated into materials science, exhibiting great potential for optimizing chemical synthesis and structural regulation, thereby accelerating intelligent design and efficient exploration of novel materials. Carbon-based emitters (CBEs), as emerging functional materials, have attracted considerable attention due to their tunable photoluminescence properties, abundant precursor sources, and structural versatility. To date, extensive research on CBEs has generated a substantial data foundation, laying the groundwork for ML driven structural screening and property prediction. Although the application of ML in CBEs remains in its early stages compared to that of inorganic semiconductors and organic-inorganic hybrid perovskites, its potential to accelerate material screening and reveal structure-property relationships is becoming increasingly evident. Given the transformative role ML has played in other functional materials, it is expected to drive a paradigm shift in CBEs research from conventional trial-and-error approaches to data-driven, intelligence-guided design, substantially accelerating material discovery and expanding their functional applications. Therefore, this review systematically summarizes recent advances in ML applications in CBEs, focusing on the general ML workflow, property prediction and structural design strategies for organic small-molecules and carbon dots (CDs), and their applications in organic light-emitting diodes (OLEDs), quantum dot light-emitting diodes (QLEDs), information encryption, biomedicine, and sensing. Finally, this review discusses the key challenges currently facing the field and offers perspectives on future directions for ML-driven CBEs research, aiming to provide guidance for the rational design and efficient development of such materials.
- 28 May 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Theory and Computation
- Vol. 22 (11) , 5399-5410
Fewest-Switches Surface Hopping with Combined Deep Learning Potential and Long Short-Term Memory Network Propagator for Simulating Realistic Photochemical Processes
Fewest-switches surface hopping (FSSH) is the most popular method for simulating photochemical processes of molecular systems. Recently, we have constructed long short-term memory (LSTM) networks as a propagator for electronic subsystems in FSSH dynamics simulations. The collective results on Tully’s three models have been reproduced satisfactorily. In the present work, we developed an extended LSTM-FSSH framework to simulate realistic photochemical reactions. The input features of LSTM as well as the training procedure are redesigned to represent high-dimensional nuclear degrees of freedom in an effective manner. Equivariant neural networks are integrated with LSTM to build adiabatic potential energy surfaces in ground and excited states. Photoisomerizations of CH2NH and azobenzene are simulated, showing that our newly proposed LSTM-FSSH method can produce excited-state lifetimes and product yields accurately in comparison with conventional FSSH simulations as a reference. Only 10 reference trajectories are required for training LSTM networks, and then a trajectory ensemble can be generated with very efficient LSTM-FSSH dynamic simulations to obtain collective results.
- 28 May 2026
- Published by American Chemical Society (ACS) in Journal of Chemical Theory and Computation
- Vol. 22 (11) , 5399-5410
Fewest-Switches Surface Hopping with Combined Deep Learning Potential and Long Short-Term Memory Network Propagator for Simulating Realistic Photochemical Processes
Fewest-switches surface hopping (FSSH) is the most popular method for simulating photochemical processes of molecular systems. Recently, we have constructed long short-term memory (LSTM) networks as a propagator for electronic subsystems in FSSH dynamics simulations. The collective results on Tully’s three models have been reproduced satisfactorily. In the present work, we developed an extended LSTM-FSSH framework to simulate realistic photochemical reactions. The input features of LSTM as well as the training procedure are redesigned to represent high-dimensional nuclear degrees of freedom in an effective manner. Equivariant neural networks are integrated with LSTM to build adiabatic potential energy surfaces in ground and excited states. Photoisomerizations of CH2NH and azobenzene are simulated, showing that our newly proposed LSTM-FSSH method can produce excited-state lifetimes and product yields accurately in comparison with conventional FSSH simulations as a reference. Only 10 reference trajectories are required for training LSTM networks, and then a trajectory ensemble can be generated with very efficient LSTM-FSSH dynamic simulations to obtain collective results.
- 27 May 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 17 (23) , 6463-6471
An Integrated Machine Learning Workflow Based on Deep Generative Models for Discovery of Inorganic Nonlinear Optical Crystals
The discovery of nonlinear optical (NLO) materials for deep-ultraviolet (DUV) and mid-infrared (MIR) applications remains a significant challenge because of stringent structural and property requirements. While machine learning (ML) has improved the prediction efficiency of NLO properties, its scope is still limited by existing data sets, resulting in insufficient exploration in uncharted chemical spaces. Here, we develop an integrated workflow based on deep generative models to overcome this challenge. Unlike most ML-based property predictors that require target labels collected from existing NLO crystals, the reference values of NLO properties are no longer necessary for training generative models. Thousands of generated structures are subsequently filtered by ML predictors and verified by first-principles calculations. Ultimately, 27 and 13 potential DUV and MIR NLO materials were identified, respectively. These crystals not only satisfy the requirements for NLO applications in the DUV or MIR spectrum but also showcase chemical compositional and structural diversity, paving a new way for the discovery of new NLO-active structural units.
- 27 May 2026
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 17 (23) , 6463-6471
An Integrated Machine Learning Workflow Based on Deep Generative Models for Discovery of Inorganic Nonlinear Optical Crystals
The discovery of nonlinear optical (NLO) materials for deep-ultraviolet (DUV) and mid-infrared (MIR) applications remains a significant challenge because of stringent structural and property requirements. While machine learning (ML) has improved the prediction efficiency of NLO properties, its scope is still limited by existing data sets, resulting in insufficient exploration in uncharted chemical spaces. Here, we develop an integrated workflow based on deep generative models to overcome this challenge. Unlike most ML-based property predictors that require target labels collected from existing NLO crystals, the reference values of NLO properties are no longer necessary for training generative models. Thousands of generated structures are subsequently filtered by ML predictors and verified by first-principles calculations. Ultimately, 27 and 13 potential DUV and MIR NLO materials were identified, respectively. These crystals not only satisfy the requirements for NLO applications in the DUV or MIR spectrum but also showcase chemical compositional and structural diversity, paving a new way for the discovery of new NLO-active structural units.
- 14 April 2026
- Published by Wiley in Angewandte Chemie
- Vol. 138 (22)
Solid‐State Red Carbon Quantum Frameworks With Narrowband Thermally Activated Delayed Fluorescence for Undoped LEDs With High‐Temperature Operational Stability
High color purity and efficiency of red light‐emitting diodes (LEDs) are essential for true‐color displays. Solution‐processable undoped emission layers represent the ideal device architecture to enhance operational lifetime and reduce fabrication complexity. However, it remains challenging to achieve narrowband emission in solid‐state, which requires reducing vibrational coupling and suppressing the π–π stacking interactions between aromatic structures. Herein, we developed solid‐state red carbon quantum frameworks (SSR‐CQFs) that exhibit narrowband thermally activated delayed fluorescence (TADF) with a peak at 635 nm, a small full width at half maximum of 39 nm, and a high photoluminescence quantum yield of 64% in neat films. This exceptional optical performance originates from the designed undulating two‐dimensional framework of SSR‐CQFs, in which chromophores are covalently interconnected by flexible alkyl chains. The structure simultaneously confines π‐electrons within individual chromophore units, prevents charge‐transfer state formation, and suppresses interlayer π–π stacking. Consequently, solution‐processed undoped LEDs based on SSR‐CQFs achieve high‐color‐purity red emission (CIE: 0.658, 0.326) with a maximum external quantum efficiency (EQE max ) of 8.04%, and demonstrate exceptional stability ( T 85 of 116 h) under high‐temperature operation. This paper presents a novel strategy to achieve solution‐processable undoped carbon‐based LEDs that achieve combination of narrowband emission, high efficiency, and high operational stability.
- 14 April 2026
- Published by Wiley in Angewandte Chemie
- Vol. 138 (22)
Solid‐State Red Carbon Quantum Frameworks With Narrowband Thermally Activated Delayed Fluorescence for Undoped LEDs With High‐Temperature Operational Stability
High color purity and efficiency of red light‐emitting diodes (LEDs) are essential for true‐color displays. Solution‐processable undoped emission layers represent the ideal device architecture to enhance operational lifetime and reduce fabrication complexity. However, it remains challenging to achieve narrowband emission in solid‐state, which requires reducing vibrational coupling and suppressing the π–π stacking interactions between aromatic structures. Herein, we developed solid‐state red carbon quantum frameworks (SSR‐CQFs) that exhibit narrowband thermally activated delayed fluorescence (TADF) with a peak at 635 nm, a small full width at half maximum of 39 nm, and a high photoluminescence quantum yield of 64% in neat films. This exceptional optical performance originates from the designed undulating two‐dimensional framework of SSR‐CQFs, in which chromophores are covalently interconnected by flexible alkyl chains. The structure simultaneously confines π‐electrons within individual chromophore units, prevents charge‐transfer state formation, and suppresses interlayer π–π stacking. Consequently, solution‐processed undoped LEDs based on SSR‐CQFs achieve high‐color‐purity red emission (CIE: 0.658, 0.326) with a maximum external quantum efficiency (EQE max ) of 8.04%, and demonstrate exceptional stability ( T 85 of 116 h) under high‐temperature operation. This paper presents a novel strategy to achieve solution‐processable undoped carbon‐based LEDs that achieve combination of narrowband emission, high efficiency, and high operational stability.
- 13 April 2026
- Published by Wiley in Angewandte Chemie International Edition
- Vol. 65 (22) , e2568667
Solid‐State Red Carbon Quantum Frameworks With Narrowband Thermally Activated Delayed Fluorescence for Undoped LEDs With High‐Temperature Operational Stability
High color purity and efficiency of red light‐emitting diodes (LEDs) are essential for true‐color displays. Solution‐processable undoped emission layers represent the ideal device architecture to enhance operational lifetime and reduce fabrication complexity. However, it remains challenging to achieve narrowband emission in solid‐state, which requires reducing vibrational coupling and suppressing the π–π stacking interactions between aromatic structures. Herein, we developed solid‐state red carbon quantum frameworks (SSR‐CQFs) that exhibit narrowband thermally activated delayed fluorescence (TADF) with a peak at 635 nm, a small full width at half maximum of 39 nm, and a high photoluminescence quantum yield of 64% in neat films. This exceptional optical performance originates from the designed undulating two‐dimensional framework of SSR‐CQFs, in which chromophores are covalently interconnected by flexible alkyl chains. The structure simultaneously confines π‐electrons within individual chromophore units, prevents charge‐transfer state formation, and suppresses interlayer π–π stacking. Consequently, solution‐processed undoped LEDs based on SSR‐CQFs achieve high‐color‐purity red emission (CIE: 0.658, 0.326) with a maximum external quantum efficiency (EQE max ) of 8.04%, and demonstrate exceptional stability ( T 85 of 116 h) under high‐temperature operation. This paper presents a novel strategy to achieve solution‐processable undoped carbon‐based LEDs that achieve combination of narrowband emission, high efficiency, and high operational stability.
- 13 April 2026
- Published by Wiley in Angewandte Chemie International Edition
- Vol. 65 (22) , e2568667
Solid‐State Red Carbon Quantum Frameworks With Narrowband Thermally Activated Delayed Fluorescence for Undoped LEDs With High‐Temperature Operational Stability
High color purity and efficiency of red light‐emitting diodes (LEDs) are essential for true‐color displays. Solution‐processable undoped emission layers represent the ideal device architecture to enhance operational lifetime and reduce fabrication complexity. However, it remains challenging to achieve narrowband emission in solid‐state, which requires reducing vibrational coupling and suppressing the π–π stacking interactions between aromatic structures. Herein, we developed solid‐state red carbon quantum frameworks (SSR‐CQFs) that exhibit narrowband thermally activated delayed fluorescence (TADF) with a peak at 635 nm, a small full width at half maximum of 39 nm, and a high photoluminescence quantum yield of 64% in neat films. This exceptional optical performance originates from the designed undulating two‐dimensional framework of SSR‐CQFs, in which chromophores are covalently interconnected by flexible alkyl chains. The structure simultaneously confines π‐electrons within individual chromophore units, prevents charge‐transfer state formation, and suppresses interlayer π–π stacking. Consequently, solution‐processed undoped LEDs based on SSR‐CQFs achieve high‐color‐purity red emission (CIE: 0.658, 0.326) with a maximum external quantum efficiency (EQE max ) of 8.04%, and demonstrate exceptional stability ( T 85 of 116 h) under high‐temperature operation. This paper presents a novel strategy to achieve solution‐processable undoped carbon‐based LEDs that achieve combination of narrowband emission, high efficiency, and high operational stability.
- 8 April 2026
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 164 (14)
Node transfer for multi-fidelity and multimodal machine learning for predicting experimental bandgaps
Bandgap is a key property of materials. In recent years, machine learning has become a powerful tool to predict the experimental bandgaps of compounds before synthesis, but there is still much room for improving the prediction accuracy. Here, we build a machine learning framework that consists of multi-fidelity and multimodal learning models to integrate heterogeneous data sources obtained from first-principle calculations and x-ray diffraction spectra. A new information-fusion strategy named node transfer is proposed. Compared to the widely used Δ-learning strategy, it naturally extends two-fidelity to multi-fidelity learning and facilitates heterogeneous multimodal integration. Node transfer consistently outperforms Δ-learning across two-fidelity, multi-fidelity, and multimodal benchmarks under fine-tuning. The best model involves XRD-based descriptors and encoded descriptors pre-trained based on four computational datasets using different functionals. It achieves a mean absolute error of 0.258 eV, a 26.3% reduction vs the single-fidelity baseline of 0.350 eV. In all prediction tasks, only the chemical composition of the crystal is required as input for the constructed machine learning models, which is free of structural information and, therefore, applicable to materials design before experiments or first-principle calculations.
- 8 April 2026
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 164 (14)
Node transfer for multi-fidelity and multimodal machine learning for predicting experimental bandgaps
Bandgap is a key property of materials. In recent years, machine learning has become a powerful tool to predict the experimental bandgaps of compounds before synthesis, but there is still much room for improving the prediction accuracy. Here, we build a machine learning framework that consists of multi-fidelity and multimodal learning models to integrate heterogeneous data sources obtained from first-principle calculations and x-ray diffraction spectra. A new information-fusion strategy named node transfer is proposed. Compared to the widely used Δ-learning strategy, it naturally extends two-fidelity to multi-fidelity learning and facilitates heterogeneous multimodal integration. Node transfer consistently outperforms Δ-learning across two-fidelity, multi-fidelity, and multimodal benchmarks under fine-tuning. The best model involves XRD-based descriptors and encoded descriptors pre-trained based on four computational datasets using different functionals. It achieves a mean absolute error of 0.258 eV, a 26.3% reduction vs the single-fidelity baseline of 0.350 eV. In all prediction tasks, only the chemical composition of the crystal is required as input for the constructed machine learning models, which is free of structural information and, therefore, applicable to materials design before experiments or first-principle calculations.
- Open Access
- 2 April 2026
- Published by Wiley in Chemistry–Methods
- Vol. 6 (4)
A Two‐Stage Machine Learning Framework With Physically Inspired Descriptors for Efficient Prediction of Hydration Free Energies
Hydration free energy is essential to understanding stability and reactivity in aqueous environments. Recently, we reported a machine learning model Mole2Solv to predict experimental hydration free energies of molecules ( J. Phys. Chem. Lett. , 2023, 14, 1877). Physically inspired descriptors such as electrostatic potential distribution around solute molecules were proposed instead of chemistry‐specific input features, thus improving transferability of the prediction model when it meets new compounds. However, most of these descriptors require ab initio calculations on solute molecules in vacuum, which seriously hinders its application on large‐scale molecular solvation modeling and high‐throughput screening. To address this limitation, we propose a two‐stage framework that combines Mole2Solv with DeepMoleNet, a multilevel attention‐based graph neural network developed by Ma and coworkers ( J. Chem. Inf. Model. , 2021, 61, 1066). Using the DeepMoleNet, 13 physically inspired descriptors can be predicted based on chemical structures of solutes and then applied to the subsequent Mole2Solv regression model. Expensive electronic structure calculation is no longer necessary for generating descriptors. In comparison with our previous model, the proposed two‐stage framework keeps the prediction accuracy while significantly reduces the computational cost. Additionally, the transferability of physically inspired descriptors to unseen chemical structures can be well preserved within the two‐stage framework.
- 2 April 2026
- Published by Wiley in Chemistry–Methods
- Vol. 6 (4)
Open AccessA Two‐Stage Machine Learning Framework With Physically Inspired Descriptors for Efficient Prediction of Hydration Free Energies
Hydration free energy is essential to understanding stability and reactivity in aqueous environments. Recently, we reported a machine learning model Mole2Solv to predict experimental hydration free energies of molecules ( J. Phys. Chem. Lett. , 2023, 14, 1877). Physically inspired descriptors such as electrostatic potential distribution around solute molecules were proposed instead of chemistry‐specific input features, thus improving transferability of the prediction model when it meets new compounds. However, most of these descriptors require ab initio calculations on solute molecules in vacuum, which seriously hinders its application on large‐scale molecular solvation modeling and high‐throughput screening. To address this limitation, we propose a two‐stage framework that combines Mole2Solv with DeepMoleNet, a multilevel attention‐based graph neural network developed by Ma and coworkers ( J. Chem. Inf. Model. , 2021, 61, 1066). Using the DeepMoleNet, 13 physically inspired descriptors can be predicted based on chemical structures of solutes and then applied to the subsequent Mole2Solv regression model. Expensive electronic structure calculation is no longer necessary for generating descriptors. In comparison with our previous model, the proposed two‐stage framework keeps the prediction accuracy while significantly reduces the computational cost. Additionally, the transferability of physically inspired descriptors to unseen chemical structures can be well preserved within the two‐stage framework.
- Open Access
Fewest-Switches Surface Hopping with Combined Deep Learning Potential and Long Short-Term Memory Network Propagator for Simulating Realistic Photochemical Processes
Fewest-switches surface hopping (FSSH) is the most popular method for simulating photochemical processes of molecular systems. Recently, we have constructed long short-term memory (LSTM) networks as a propagator for electronic subsystems in FSSH dynamics simulations. The collective results on Tully's three models have been reproduced satisfactorily. In the present work, we develop an extended LSTM-FSSH framework to simulate realistic photochemical reactions. The input features of LSTM as well as the training procedure are redesigned to represent high-dimensional nuclear degrees of freedom in an effective way. Equivariant neural networks are integrated with LSTM to build adiabatic potential energy surfaces in ground and excited states. Photoisomerizations of $\mathrm{CH_2NH}$ and azobenzene are simulated, showing that our new proposed LSTM-FSSH method can produce excited-state lifetimes and product yields accurately in comparison with conventional FSSH simulations as reference. Only 10 reference trajectories are required for training LSTM networks, and then a trajectory ensemble can be generated with very efficient LSTM-FSSH dynamics simulations to obtain collective results.Open AccessFewest-Switches Surface Hopping with Combined Deep Learning Potential and Long Short-Term Memory Network Propagator for Simulating Realistic Photochemical Processes
Fewest-switches surface hopping (FSSH) is the most popular method for simulating photochemical processes of molecular systems. Recently, we have constructed long short-term memory (LSTM) networks as a propagator for electronic subsystems in FSSH dynamics simulations. The collective results on Tully's three models have been reproduced satisfactorily. In the present work, we develop an extended LSTM-FSSH framework to simulate realistic photochemical reactions. The input features of LSTM as well as the training procedure are redesigned to represent high-dimensional nuclear degrees of freedom in an effective way. Equivariant neural networks are integrated with LSTM to build adiabatic potential energy surfaces in ground and excited states. Photoisomerizations of $\mathrm{CH_2NH}$ and azobenzene are simulated, showing that our new proposed LSTM-FSSH method can produce excited-state lifetimes and product yields accurately in comparison with conventional FSSH simulations as reference. Only 10 reference trajectories are required for training LSTM networks, and then a trajectory ensemble can be generated with very efficient LSTM-FSSH dynamics simulations to obtain collective results. - 17 September 2025
- Published by Springer Nature in Chemical Research in Chinese Universities
- Vol. 41 (5) , 1014-1020
Darwin4Matter: A Platform Integrating Machine Learning and Quantum Chemistry for New Materials Design
- 17 September 2025
- Published by Springer Nature in Chemical Research in Chinese Universities
- Vol. 41 (5) , 1014-1020
Darwin4Matter: A Platform Integrating Machine Learning and Quantum Chemistry for New Materials Design