QD3SET-1: A Database with Quantum Dissipative Dynamics Data Sets

In a recent article published in Frontiers in Physics, we introduce QD3SET-1, a database consisting of 8 data sets that provide the time-evolved population and coherence dynamics for two widely studied systems: the so-called spin-boson model and FMO complex. The methodologies employed for dynamics propagation are the hierarchical equations of motion (HEOM) approach and the Locally Thermalized Lindblad Equation of Motion (LTLME). The primary objective behind the release of the QD3SET-1 database is to provide researchers a valuable resource for the development, testing, and validation of their approaches, whether rooted in machine learning or other non-machine learning methodologies.

Simulating the inherently quantum-mechanical dynamics of charge, energy, and coherence transfer in condensed-phase systems presents several challenges. The exponential scaling of computational cost limits the feasibility of performing quantum-mechanically exact simulations for complex systems. Approximations become necessary, and this is where machine learning (ML) methods come into play. By leveraging ML approaches, researchers can overcome the limitations of traditional quantum dynamics methods and achieve accurate results more efficiently. [1, 2, 3, 4]

As we know, accelerating progress in machine learning (ML) requires access to large and diverse datasets for training purposes. In the realm of quantum chemistry, numerous open-source databases exist (QM-sym, QM7, QM7b, MD17, GDB-13, QM9, VIB5, and WS22), which have proven to be invaluable resources, enabling researchers to directly utilize existing data for their studies. However, in the domain of quantum dissipative dynamics, the availability of such databases has been limited or non-existent. This poses a significant challenge, as generating the required data for training ML models in quantum dissipative dynamics can be a time-consuming and expensive process.

Recognizing this gap, we are excited to introduce QD3SET-1, an innovative database specifically designed to provide quantum dissipative dynamics data sets.

Introducing the QD3SET-1 Database

The QD3SET-1 database comprises eight distinct data sets that provide the time-evolved population and coherence dynamics of two important systems: the spin-boson (SB) model and the Fenna–Matthews–Olson (FMO) light-harvesting complex.

The SB data set within QD3SET-1 contains 1000 trajectories generated using the hierarchical equations of motion (HEOM) method. These trajectories explore a wide range of system-bath coupling strengths (λ), characteristic frequencies (γ), and inverse temperatures (β). Both symmetric and asymmetric cases are considered, ensuring comprehensive coverage. Each trajectory is propagated with a time step of 0.05 and extended up to tΔ=20, where Δ represents the tunneling matrix element.

Moving on to the FMO data sets, QD3SET-1 encompasses 7 data sets for various FMO Hamiltonians and initial excitations, providing researchers with ample opportunities to investigate different parameters and scenarios. Six out of the seven FMO data sets are generated using the Locally Thermalized Lindblad Equation of Motion (LTLME) approach. For each case with an initial excitation, 500 trajectories are generated, systematically varying the system-bath coupling strength (λ), characteristic frequency (γ), and temperature (T) based on farthest point sampling. These trajectories are propagated for 50 ps with a time step of 5 fs. One of FMO data sets stands has trajectories propagated even up to 1 ns.

In addition to LTLME, one FMO data set within QD3SET-1 is generated using the hierarchical equations of motion (HEOM) method. A total of 879 trajectories are generated for a single excitation case, with a propagation time of 2 ps and a time step of 0.1 fs.

Applications of the QD3SET-1 Database

The availability of QD3SET-1 database offers several benefits to researchers. First and foremost, it enables the development and benchmarking of new simulation methodologies, both physics-based and machine learning-based. The data sets cover a diverse scenario of two commonly used quantum systems, providing researchers with a valuable resource for testing and validating their approaches.

In addition, the QD3SET-1 database has already been utilized in previous studies to develop and benchmark machine learning approaches for quantum dissipative dynamics.[Ullah and Dral (2021), Ullah and Dral (2022), Rodríguez et al. (2022)] These studies have demonstrated the potential of ML techniques in accelerating simulations and predicting system behaviors accurately.

Furthermore, the availability of coherently formatted data, metadata, and extraction scripts in the form of QD3SET-1 promotes collaboration and accelerates innovation in the field. As the field of quantum dynamics continues to evolve, databases like QD3SET-1 will play a crucial role in driving scientific progress and enabling new discoveries.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.