How Sophisticated Are Neural Networks Needed to Predict Long-Term Nonadiabatic Dynamics? DOI Creative Commons
Hao Zeng,

Yitian Kou,

Xiang Sun

et al.

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(22), P. 9832 - 9848

Published: Nov. 14, 2024

Nonadiabatic dynamics is key for understanding solar energy conversion and photochemical processes in condensed phases. This often involves the non-Markovian of reduced density matrix open quantum systems, where knowledge system's prior states necessary to predict its future behavior. In this study, we explore time-series machine learning methods predicting long-time nonadiabatic based on short-time input data, comparing these with physics-based transfer tensor method (TTM). To understand impact memory time approaches, demonstrate that can be represented as a linear map within Nakajima-Zwanzig generalized master equation framework. We further propose practical estimate effective time, given tolerance, propagation. Our predictive models are applied various physical including spin-boson models, multistate harmonic (MSH) Ohmic spectral densities realistic organic photovoltaic system composed carotenoid-porphyrin-fullerene triad dissolved tetrahydrofuran. Results indicate simple linear-mapping fully connected neural network (FCN) outperforms more complicated nonlinear-mapping networks gated recurrent unit (GRU) convolutional network/long short-term (CNN-LSTM) systems short times, such MSH models. Conversely, nonlinear CNN-LSTM GRU yield higher accuracy characterized by long times. These findings offer valuable insights into role dynamics, providing guidance application complex chemical systems.

Language: Английский

Radiative pumping vs vibrational relaxation of molecular polaritons: a bosonic mapping approach DOI Creative Commons
Juan B. Pérez-Sánchez, Joel Yuen-Zhou

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: April 2, 2025

Language: Английский

Citations

0
Two-dimensional coherent spectrum of high-spin models via a quantum computing approach DOI Creative Commons
Martin Mootz, Peter P. Orth, Chuankun Huang

et al.

Quantum Science and Technology, Journal Year: 2024, Volume and Issue: 9(3), P. 035054 - 035054

Published: June 13, 2024

Abstract We present and benchmark a quantum computing approach to calculate the two-dimensional coherent spectrum (2DCS) of high-spin models. Our is based on simulating their real-time dynamics in presence several magnetic field pulses, which are spaced time. utilize adaptive variational simulation algorithm for study due its compact circuits, enables simulations over sufficiently long times achieve required resolution frequency space. Specifically, we consider an antiferromagnetic spin model that incorporates Dzyaloshinskii-Moriya interactions single-ion anisotropy. The obtained 2DCS spectra exhibit distinct peaks at multiples magnon frequency, arising from transitions between different eigenstates unperturbed Hamiltonian. By comparing one-dimensional with 2DCS, demonstrate provides higher energy spectrum. further investigate how resources scale magnitude using two binary encodings operators: standard encoding Gray code. At low fields both require comparable resources, but larger strengths code advantageous. Numerical models increasing number sites indicate polynomial system-size scaling resources. Lastly, compare numerical experimental results rare-earth orthoferrite system. observed strength magnonic high-harmonic generation signals aligns well data, showing significant improvement corresponding mean-field results.

Language: Английский

Citations

3
Unitary collapse of Schrödinger's cat state DOI
Pavel Stránský, Pavel Cejnar, Radim Filip

et al.

Physical review. A/Physical review, A, Journal Year: 2024, Volume and Issue: 110(3)

Published: Sept. 19, 2024

Language: Английский

Citations

1
How Sophisticated Are Neural Networks Needed to Predict Long-Term Nonadiabatic Dynamics? DOI Creative Commons
Hao Zeng,

Yitian Kou,

Xiang Sun

et al.

Journal of Chemical Theory and Computation, Journal Year: 2024, Volume and Issue: 20(22), P. 9832 - 9848

Published: Nov. 14, 2024

Nonadiabatic dynamics is key for understanding solar energy conversion and photochemical processes in condensed phases. This often involves the non-Markovian of reduced density matrix open quantum systems, where knowledge system's prior states necessary to predict its future behavior. In this study, we explore time-series machine learning methods predicting long-time nonadiabatic based on short-time input data, comparing these with physics-based transfer tensor method (TTM). To understand impact memory time approaches, demonstrate that can be represented as a linear map within Nakajima-Zwanzig generalized master equation framework. We further propose practical estimate effective time, given tolerance, propagation. Our predictive models are applied various physical including spin-boson models, multistate harmonic (MSH) Ohmic spectral densities realistic organic photovoltaic system composed carotenoid-porphyrin-fullerene triad dissolved tetrahydrofuran. Results indicate simple linear-mapping fully connected neural network (FCN) outperforms more complicated nonlinear-mapping networks gated recurrent unit (GRU) convolutional network/long short-term (CNN-LSTM) systems short times, such MSH models. Conversely, nonlinear CNN-LSTM GRU yield higher accuracy characterized by long times. These findings offer valuable insights into role dynamics, providing guidance application complex chemical systems.

Language: Английский

Citations

1