An Efficient Variational Multiscale Fractional-Step Method for Incompressible Flow with Consistent Boundary Conditions

  • Wu, Peishuo (Peking university)
  • Zhu, Chi (Peking university)

Please login to view abstract download link

The variational multiscale (VMS) method [1] was developed to mitigate pressure and advection instabilities in the standard Galerkin formulation of the finite element method. It has since been successfully extended from the monolithic setup to the fractional-step setup [2], demonstrating superior computational efficiency. However, applying this method to complex problems, such as cardiovascular flow, remains challenging, including the treatment of open/traction boundary conditions (BCs) and the attainment of stable solutions with linear elements under these BCs. Adopting linear elements requires careful design of stabilization terms, which has been investigated only under simple Dirichlet BCs in a fractional-step setup [3]. On the other hand, consistent BCs, free of numerical boundary layers, can be achieved using the rotational pressure-correction scheme (RPCS) [4]. Nevertheless, RPCS has been investigated only in the context of inf-sub stable element pairs (e.g., P2/P1 elements) [5]. Therefore, there is a need to develop a VMS-based fractional-step method for linear elements with consistent BCs. In this study, we propose an efficient VMS fractional-step method for incompressible flow that accurately handles a mixture of Dirichlet and open BCs. We adopt the RPCS formula and derive new VMS terms for the Poisson and mass equations. The proposed scheme is verified through several numerical experiments that solve unsteady, viscous flow problems at various Reynolds numbers. Stable and accurate solutions for the pressure and velocity fields are obtained using P1/P1 elements. A quantitative convergence study shows that, compared with the traditional pressure-stabilized Petrov–Galerkin scheme with P1/P1 elements, the proposed fractional-step scheme exhibits superior spatial convergence and equivalent temporal convergence. In all numerical cases, the proposed scheme is nearly twice as computationally efficient as the monolithic scheme while yielding results that agree with published benchmarks.