Abstract:Molecular Dynamics (MD) simulation is a powerful tool for understanding the dynamics and structure of matter. Since the resolution of MD is atomic-scale, achieving long time-scale simulations with femtosecond integration is very expensive. In each MD step, numerous iterative computations are performed to calculate energy based on different types of interaction and their corresponding spatial gradients. These repetitive computations can be learned and surrogated by a deep learning model like a Graph Neural Network (GNN). In this work, we developed a GNN Accelerated Molecular Dynamics (GAMD) model that directly predicts forces given the state of the system (atom positions, atom types), bypassing the evaluation of potential energy. By training the GNN on a variety of data sources (simulation data derived from classical MD and density functional theory), we show that GAMD can predict the dynamics of two typical molecular systems, Lennard-Jones system and Water system, in the NVT ensemble with velocities regulated by a thermostat. We further show that GAMD's learning and inference are agnostic to the scale, where it can scale to much larger systems at test time. We also perform a comprehensive benchmark test comparing our implementation of GAMD to production-level MD softwares, showing GAMD's competitive performance on the large-scale simulation.
What problem does this paper attempt to address?
The problems that this paper attempts to solve are the two major limitations in Molecular Dynamics (MD) simulations: computational complexity and accuracy issues.
1. **Computational Complexity**:
- Due to its atomic - level resolution, molecular dynamics simulations require time integration at the femtosecond level, which makes long - time - scale simulations very costly. In each MD step, a large number of iterative calculations are required to calculate the energy based on different types of interactions and their corresponding spatial gradients. These repetitive calculations can be learned and replaced by deep - learning models such as Graph Neural Networks (GNNs).
- In traditional MD simulations, each iterative step requires a large amount of calculation, including calculating the forces acting on each particle. These forces are obtained by adding up various bonding or non - bonding potential energies in the system. Depending on the type and scale of the simulation, especially when calculating the force on each atom as the negative gradient of the empirical potential energy, these calculations are particularly time - consuming.
2. **Accuracy Issues**:
- Classical MD simulations rely on empirical potential functions to calculate forces, and these potential functions have limitations in accuracy and generalization ability when describing complex inter - atomic interactions. Finding an appropriate function form that can meet the accuracy requirements is a challenge.
- Although ab - initio methods such as Density Functional Theory (DFT) are more accurate, they are too computationally expensive for large - scale many - body systems.
To solve these problems, the author has developed a Graph Neural Network - Accelerated Molecular Dynamics model (GNN Accelerated Molecular Dynamics, GAMD). This model directly predicts forces based on the state of the system (atomic positions, atomic types) without evaluating the potential energy. In this way, GAMD avoids the computational bottleneck of calculating the spatial derivative of the potential energy in traditional MD simulations and demonstrates the ability to predict dynamics for typical molecular systems (such as Lennard - Jones systems and water systems) in the NVT ensemble. In addition, GAMD also shows good scalability. It can be extended to larger systems at test time and has performance comparable to production - level MD software in large - scale simulations.
Expressed in a formula, the core idea of GAMD is:
\[
\text{GAMD} \to f_i = F(p, q_i, b_{ij})
\]
where \( f_i \) is the predicted force of the \( i \) - th atom, \( p \) is the atomic type, \( q_i \) is the atomic position, and \( b_{ij} \) is the bond information between atoms. By training the GNN, GAMD can directly learn from the input features and predict forces, thereby accelerating molecular dynamics simulations.