Ceci n'est pas un committor, yet it samples like one: Efficient sampling via approximated committor functions

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Atomistic simulations are crucial for studying chemical reactions but face challenges with rare events and kinetic bottlenecks.
• Enhanced sampling methods are needed to overcome these limitations and efficiently explore reaction pathways.

Purpose of the Study:

• To develop a computationally efficient enhanced sampling method for atomistic simulations.
• To address the limitations of existing committor function-based approaches by simplifying the learning criterion.

Main Methods:

• Introduced a simplified learning criterion for machine-learning the committor function in atomistic simulations.
• Formulated the learning criterion entirely in the descriptor space, avoiding costly coordinate gradients.
• Combined a transition-state-oriented bias potential with a metadynamics-like bias along a committor-based collective variable.

Main Results:

• The simplified learning criterion significantly reduces computational costs compared to the original formulation.
• The method retains robust sampling performance despite not formally targeting the exact committor.
• Enables the study of processes previously deemed computationally unfeasible.

Conclusions:

• The proposed simplified learning criterion offers a practical and efficient alternative for enhanced sampling in atomistic simulations.
• This advancement facilitates the investigation of complex reactive processes with reduced computational burden.
• The method holds promise for broader applications in computational chemistry and materials science.