Study of phase-field lattice Boltzmann models based on the conservative Allen-Cahn equation

Conservative phase-field (CPF) equations based on the Allen-Cahn model for interface tracking in multiphase flows have become more popular in recent years, especially in the lattice-Boltzmann (LB) community. This is largely due to their simplicity and improved efficiency and accuracy over their Cahn-Hilliard-based counterparts. Additionally, the improved locality of the resulting LB equation (LBE) for the CPF models makes them more ideal candidates for LB simulation of multiphase flows on nonuniform grids, particularly within an adaptive-mesh refinement framework and massively parallel implementation. In this regard, some modifications-intended as improvements-have been made to the original CPF-LBE proposed by Geier et al. [Phys. Rev. E 91, 063309 (2015)] which require further examination. The goal of the present study is to conduct a comparative investigation into the differences between the original CPF model proposed by Geier et al. [Phys. Rev. E 91, 063309 (2015)] and the so-called improvements proposed by Ren et al. [Phys. Rev. E 94, 023311 (2016)] and Wang et al. [Phys. Rev. E 94, 033304 (2016)]. Using the Chapman-Enskog analysis, we provide a detailed derivation of the governing equations in each model and then examine the efficacy of the above-mentioned models for some benchmark problems. Several test cases have been designed to study different configurations ranging from basic yet informative flows to more complex flow fields, and the results are compared with finite-difference simulations. Furthermore, as a development of the previously proposed CPF-LBE model, axisymmetric formulations for the proposed model by Geier et al. [Phys. Rev. E 91, 063309 (2015)] are derived and presented. Finally, two benchmark problems are designed to compare the proposed axisymmetric model with the analytical solution and previous work. We find that the accuracy of the model for interface tracking is roughly similar for different models at high viscosity ratios, high density ratios, and relatively high Reynolds numbers, while the original CFP-LBE without the additional time-dependent terms outperforms the so-called improved models in terms of efficiency, particularly on distributed parallel machines.