Convergence analysis of an efficient scheme for the steady state second grade fluid model (Q6591013)

The authors consider second-grade fluids for which the Cauchy stress tensor \( T\) is a polynomial of degree two in the first two Rivlin-Ericksen tensors \( T_{1}=\nabla u+\nabla u^{t}\) and \(T_{2}=\frac{dT_{1}}{dt}+T_{1}\nabla u+\nabla u^{t}T_{1}\), where \(u\) is the velocity of the fluid and \(\frac{d}{dt }=\partial _{t}+u\cdot \nabla \). They write the balance of linear momentum: \( \rho \frac{du}{dt}=\operatorname{div}T+\rho f\), where \(\rho \) is the density and \(f\) an external force, and they obtain in the steady case the equations describing the motion of an incompressible second-grade fluid: \(-\nu \Delta u+\mathrm{curl}(u-(2\alpha _{1}+\alpha _{2})\Delta u)\times u-(\alpha _{1}+\alpha _{2})\Delta (u\cdot \nabla u)+2(\alpha _{1}+\alpha _{2})u\cdot \nabla (\Delta u)+\nabla p=f\), \(\operatorname{div}u=0\), where \(\nu \), \(\alpha _{1}\), \(\alpha _{2}\) are constants and \(p\) is the pressure. Because of the thermodynamical principles \(\nu \geq 0\), \(\alpha _{1}\geq 0\), \(\alpha _{2}=-\alpha _{1}=\alpha \), the equations simplify as: \(-\nu \Delta u+\mathrm{curl}(u-\alpha \Delta u)\times u+\nabla p=f\), \(\operatorname{div}u=0\), and the problem is posed in a regular domain \(\Omega \subset \mathbb{R}^{2}\). The boundary conditions \(u=0\) are imposed on \(\partial \Omega \). Introducing the new variable \(z=\mathrm{curl} (u-\alpha \Delta u)\) and a regularizing parameter \(\varepsilon \), the authors end with the system: \(-\nu \Delta u+z\times u+\nabla p=f\), \(\operatorname{div}u=0\), \(\nu z-\varepsilon \Delta uz+\alpha u\cdot \nabla z=\alpha \mathrm{curl} \ f+\nu \mathrm{curl} \ u\), for which they introduce a variational formulation in \( H_{0}^{1}(\Omega )\times L_{0}^{2}(\Omega )\times H^{1}(\Omega )\). Assuming that \(f\) belongs to \(H^{1}(\mathrm{curl})=\{f\in L^{2}(\Omega ):\mathrm{curl} \ f\in L^{2}(\Omega )\}\), this system admits a solution \((u,p,z)\in H_{0}^{1}(\Omega )\times L_{0}^{2}(\Omega )\times H^{1}(\Omega )\) satisfying uniform estimates in terms of \(\left\Vert f\right\Vert _{L^{2}(\Omega )}\) and \(\left\Vert \mathrm{curl} \ f\right\Vert _{L^{2}(\Omega )}\). Under a further hypothesis on the data, this solution is unique. This solution converges to a weak solution to the original system when the penalization parameter \(\varepsilon \) tends to zero. The authors then propose an iterative scheme to solve the system. In a first attempt, the authors compute \((u_{k+1},p_{k+1})\in H_{0}^{1}(\Omega )\times L_{0}^{2}(\Omega )\) solution to the variational formulations of the two first equations with \(z_{k}\in H^{1}(\Omega )\) known, then they compute \( z_{k+1}\in H^{1}(\Omega )\) solution to the variational formulation of the third equation with \((u_{k+1},p_{k+1})\in H_{0}^{1}(\Omega )\times L_{0}^{2}(\Omega )\). They prove the existence of a solution to this iterative scheme which is unique under some further condition on the data. They prove that this scheme converges to a solution to the variational formulation of the system. The authors finally propose a fully decoupled scheme using the Grad-Div stabilized operator, and they prove a convergence result for this scheme. In the last part of their paper, the authors present the results of numerical simulations obtained through the last scheme. In a first case, the domain \(\Omega \) is chosen as \(]0,\pi \lbrack ^{2}\) and an exact solution is known, the parameter \(\alpha \) is fixed at \(0.1\) and the viscosity \(\nu \) set to \(1\). The authors test the sensitivity with respect to the Grad-Div parameter. In the second case, the authors consider a 2D lid-driven flow with a viscosity equal to 0.1 and with different value of parameter \(\alpha \) between 10\(^{-5}\) and 5. The computational domain \( \Omega \) is the square \(]0,1[^{2}\).