Machine Learning Models Capture Plasmon Dynamics in Ag Nanoparticles.

Area of Science:

• Computational materials science
• Plasmonics
• Energy harvesting

Background:

• Hot carriers from plasmon decay in nanostructures offer potential for sustainable energy devices.
• Efficiently collecting these carriers before thermalization is a major challenge.
• Atomistic theoretical investigations are crucial but computationally expensive with traditional methods.

Purpose of the Study:

• To develop a computationally efficient method for simulating hot carrier dynamics in metallic nanostructures.
• To enable detailed analysis of plasmon dynamics in larger and more complex systems.
• To accelerate the understanding of fundamental processes in plasmon-driven hot carrier devices.

Main Methods:

• Modification of the Hierarchically Interacting Particle Neural Network (HIP-NN) for predicting plasmon dynamics.
• Utilizing real-time time-dependent density functional theory (rt-TDDFT) for training data.
• Implementing a multistep training approach to stabilize predictions over longer trajectories.
• Employing Graphics Processing Units (GPUs) for accelerated simulations.

Main Results:

• The modified HIP-NN model accurately predicts plasmon dynamics in silver (Ag) nanoparticles, agreeing well with rt-TDDFT simulations.
• The model successfully simulates trajectories up to ~25 fs for nanoparticles up to 561 atoms, including those not in the training set.
• Achieved speed-up factors of ~10^3 and ~10^4 compared to rt-TDDFT for small and large nanoparticles, respectively.
• Demonstrated accurate prediction of physical quantities like dynamic dipole moments.

Conclusions:

• Machine learning, specifically the modified HIP-NN, significantly accelerates the simulation of electron/nuclear dynamics.
• This approach overcomes the computational cost limitations of first-principles methods for large nanostructures.
• The findings highlight the potential of ML-accelerated simulations for advancing the design and understanding of plasmon-driven hot carrier energy devices.