Exploring the Reaction Network of Acetic Acid in Supercritical Water via Machine Learning Interatomic Potential.

Area of Science:

• Chemical Engineering
• Computational Chemistry
• Environmental Science

Background:

• Supercritical water oxidation (SCWO) is a promising waste treatment technology.
• Understanding SCWO's complex molecular reaction mechanisms is difficult due to extreme conditions.
• Acetic acid oxidation in SCW is a key industrial process.

Purpose of the Study:

• Compare machine learning potentials (NequIP) and reactive force fields (ReaxFF) for modeling acetic acid oxidation in SCW.
• Evaluate the accuracy of NequIP and ReaxFF in predicting reaction pathways, product distributions, and mechanisms.
• Assess the suitability of these computational methods for optimizing industrial oxidation processes.

Main Methods:

• Utilized NequIP, a machine learning potential, and ReaxFF, a reactive force field, for computational modeling.
• Simulated acetic acid oxidation in supercritical water under varying oxidant conditions (O2 and H2O2).
• Compared model predictions against experimental data for activation barriers, product distributions, and reaction pathways.

Main Results:

• NequIP accurately reproduced experimental product distributions and confirmed radical reaction mechanisms.
• ReaxFF predicted activation barriers closer to experimental values but overestimated intermediate stability and favored incomplete oxidation.
• Both models predicted enhanced reaction rates with hydrogen peroxide, but with differing impacts on specific steps.

Conclusions:

• Machine learning potentials like NequIP offer a balance of accuracy and efficiency for modeling complex SCWO reactions.
• NequIP provides superior insights into SCWO mechanisms and product formation compared to ReaxFF.
• These computational tools can guide the optimization of industrial waste treatment processes.