Reconstructing Pristine Molecular Orbitals from Scanning Tunneling Microscope Images via Artificial Intelligence...

Area of Science:

• Physical Chemistry
• Surface Science
• Computational Chemistry

Background:

• Molecular orbitals (MOs) are fundamental to chemistry, and scanning tunneling microscopy (STM) offers potential for their spatial characterization.
• High-resolution imaging of MOs using STM is challenging due to interference from functionalized tips, specifically high-angular-momentum contributions.

Purpose of the Study:

• To develop a novel method for accurate, high-resolution reconstruction of molecular orbitals from STM images.
• To overcome the limitations imposed by functionalized tips in STM imaging of MOs.

Main Methods:

• Establishment of a physics-driven deep-learning network, STM-Net, for MO reconstruction.
• Leveraging AI for image recognition and the separable characteristics of different angular momentum contributions.
• Application of the network to experimental STM data with functionalized tips.

Main Results:

• STM-Net successfully reconstructs pristine molecular orbital features from high-resolution STM images.
• The method demonstrates direct applicability to diverse experimental observations and molecular conditions.
• The framework shows adaptability to various functionalized tip states and substrates.

Conclusions:

• STM-Net provides a robust solution for accurate, high-resolution characterization of molecular orbitals.
• This physics-driven AI approach broadens the applicability of STM for fundamental chemical insights.
• The findings may lead to new applications and a deeper understanding of molecular orbitals.