Autonomous Single-Molecule Manipulation Based on Reinforcement Learning.

Area of Science:

• Nanoscience
• Molecular Engineering
• Artificial Intelligence

Background:

• Precise control of single molecules is crucial for building nanostructures atom-by-atom.
• Complex, repetitive molecular manipulation tasks are time-consuming and ideal for machine learning.
• Scanning tunneling microscopy (STM) enables manipulation of individual molecules.

Purpose of the Study:

• To develop and train a reinforcement learning (RL) algorithm for controlling single dipolar molecules using an STM.
• To investigate the physical insights into molecular movement and orientation generated by the RL algorithm.
• To understand the influence of electric field application on molecular manipulation.

Main Methods:

• A reinforcement learning algorithm was trained to control a single dipolar molecule.
• The molecule was manipulated within the electric field of an STM.
• The training process involved approximately 2250 iterations.
• The algorithm learned to guide the molecule to specific surface positions.

Main Results:

• The RL algorithm successfully learned to manipulate the molecule to target positions.
• The algorithm provided physical insights into molecular movement and orientation.
• Analysis revealed directional inhibition of molecular movement.
• Torque was found to be asymmetric around the molecule's dipole moment.

Conclusions:

• Reinforcement learning can effectively control single molecules for nanomanipulation.
• The developed algorithm provides valuable physical insights into molecular dynamics.
• Understanding directional movement constraints and asymmetric torques is key for advanced molecular control.