Permutation invariant polynomial neural network approach to fitting potential energy surfaces. III. Molecule-surface...

Area of Science:

• Computational Chemistry
• Materials Science
• Physical Chemistry

Background:

• Accurate potential energy surfaces (PESs) are crucial for understanding molecular interactions.
• Existing methods for gas-phase molecules need adaptation for molecule-surface systems.
• Neural networks offer a powerful tool for constructing complex PESs.

Purpose of the Study:

• To extend the permutation invariant polynomial-neural network (PIP-NN) method to molecule-surface interaction PESs.
• To incorporate surface translational and molecular permutation symmetries into the neural network fitting process.
• To demonstrate the accuracy and efficiency of the adapted PIP-NN method.

Main Methods:

• Developed symmetry functions that satisfy both surface translational and molecular permutation symmetries.
• Utilized low-order permutation invariant polynomials (PIPs) of primitive symmetry functions.
• Incorporated permutationally invariant cross terms to prevent oversymmetrization.
• Applied the method to fit a model PES for H2 + Cu(111) and DFT points for H2 + Ag(111).

Main Results:

• Successfully adapted the PIP-NN method for molecule-surface interaction PESs.
• Achieved accurate fitting of PESs by incorporating appropriate symmetry functions.
• Demonstrated the method's efficiency and accuracy on benchmark systems.
• Validated the approach using density functional theory calculations.

Conclusions:

• The extended PIP-NN method provides a robust framework for calculating molecule-surface interaction PESs.
• Symmetry adaptation is key to achieving high accuracy in neural network-based PES construction.
• This approach facilitates more reliable simulations of surface chemistry and catalysis.