NaCl nucleation from brine in seeded simulations: Sources of uncertainty in rate estimates

Abstract

This work reexamines seeded simulation results for NaCl nucleation from a supersaturated aqueous solution at 298.15 K and 1 bar pressure. We present a linear regression approach for analyzing seeded simulation data that provides both nucleation rates and uncertainty estimates. Our results show that rates obtained from seeded simulations rely critically on a precise driving force for the model system. The driving force vs. solute concentration curve need not exactly reproduce that of the real system, but it should accurately describe the thermodynamic properties of the model system. We also show that rate estimates depend strongly on the nucleus size metric. We show that the rate estimates systematically increase as more stringent local order parameters are used to count members of a cluster and provide tentative suggestions for appropriate clustering criteria.

Nucleation in aqueous NaCl solutions shifts from 1-step to 2-step mechanism on crossing the spinodal

“…34 The JC NaCl model has an equilibrium solubility of 3.7 mol/kg in SPC/E water at 298.15 K and 1 bar, confirmed by both chemical potential 35 and direct coexistence methods. 36 The nucleation rate for this system has been recently studied 12,37 using forward flux sampling methods, and from seeding simulations in conjunction with classical nucleation theory (CNT) 19,38 ; it was demonstrated that at modest supersaturations nucleation follows a classical one-step mechanism.…”

Seeding approach to nucleation in the NVT ensemble: The case of bubble cavitation in overstretched Lennard Jones fluids

“…A rather recent approach, named Seeding, consists in directly starting the simulation from a configuration where the nucleus of the stable phase is already formed [13][14][15][16][17]. This approach is not fully rigorous as it relies on the validity of Classical Nucleation Theory [6,18,19] and on a judicious choice for the criterion used to determine the nucleus size [20]. Despite the lack of rigor, this method has proven successful in predicting nucleation rates in crystal nucleation of hard spheres, Lennard Jones spheres, water or sodium chloride [17].…”

A generalized deep learning approach for local structure identification in molecular simulations

“…These types of simulations are used as part of the seeding method,49 which is used to predict crystal nucleation rates. The definition of the exact size of the crystal seed is one of the largest sources of uncertainty in the method 50. The test case enabled us to explore several questions regarding the behavior of the PointNet.…”