Machine learning many-body potentials for charged colloids in primitive 1:1 electrolytes.

Area of Science:

• Colloid and Interface Science
• Computational Physics
• Materials Science

Background:

• The Derjaguin-Landau-Verwey-Overbeek (DLVO) potential is standard for modeling charged particle interactions in electrolytes.
• Experiments reveal DLVO theory's limitations under strong Coulomb coupling due to nontrivial ion correlations.
• Accurate modeling requires explicit inclusion of ions, but direct simulations are computationally intensive.

Purpose of the Study:

• To develop a computationally efficient method for simulating charged colloidal suspensions.
• To overcome the slow equilibration challenge in direct ion-inclusive simulations.
• To enable large-scale studies of colloidal phase behavior.

Main Methods:

• Employed a machine-learning (ML) framework to generate density-dependent ML potentials.
• ML potentials accurately describe effective colloid interactions for given system parameters.
• Facilitated fast and large-scale simulations of charged colloids.

Main Results:

• Developed ML potentials that capture complex ion-colloid interactions.
• Achieved significantly faster simulation speeds compared to traditional methods.
• Enabled the possibility of systematic studies on colloidal phase behavior.

Conclusions:

• The ML framework provides a viable and efficient approach for simulating strongly coupled colloidal systems.
• This method opens new avenues for exploring gas-liquid and fluid-solid coexistence in charged colloids.
• Machine learning offers a powerful tool to advance the understanding of complex soft matter systems.