Computing free energy barriers for the nucleation of complex network mesophases.

A previously introduced framework to identify local order parameters (OPs) distinctive of incipient complex mesophases, such as bicontinuous network phases, is used in this work to evaluate nucleation free-energy barriers. The sampling techniques considered are the mean-first-passage-time (MFPT) method and novel variants of umbrella sampling, including Hybrid Monte Carlo (HMC) and a dual-OP-method that uses a blunter global OP for the umbrella bias while keeping record of configurations for analysis with a local OP. These methods were chosen for their ability to minimize or avoid frequent calculation of the expensive local OP, which makes their continuous on-the-fly tracking computationally very inefficient. These techniques were first validated by studying phase-transition barriers of model systems, i.e., the vapor-liquid nucleation of Lennard-Jones argon and a binary nanoparticle model. The disorder-to-order free energy barrier was then traced for the double gyroid and single diamond formed by mesoscopic bead-spring macromolecular models. The dual OP method was found to be the most robust and computationally efficient, since, unlike HMC, it does not require the expensive local OP to be computed on-the-fly, and unlike the MFPT method, it can negotiate large barriers aided by the biased sampling. The dual OP method requires, however, that a cheap global OP be identified and correlated (in a post-processing step) with the local OP that describes the structure of the critical nucleus, a process that can be aided by machine learning.