The generalized scalar auxiliary variable applied to the incompressible Boussinesq Equation

Andreas Wagner, Barbara Wohlmuth, Jan Zawallich · 2025

Read The generalized scalar auxiliary variable applied to the incompressible Boussinesq Equation by Andreas Wagner, Barbara Wohlmuth, Jan Zawallich (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-13374.txt.

Introduction

Andreas Wagner, Barbara Wohlmuth, and Jan Zawallich write a 2025 paper in numerical mathematics. Their topic is the Boussinesq equation, which models fluid flow driven by heat. The team builds a new way to step such flows forward in time with care.

The scheme pairs a generalized scalar auxiliary variable with a backward formula of second order. A special step for the extra variable keeps the method stable for any step size. The authors prove error bounds, then recast the scheme for use with finite elements. Tests with a Taylor-Hood setup back the theory and show the method can handle large three-dimensional problems in fluid flow driven by heat.

Catalog record

CREATORAndreas Wagner, Barbara Wohlmuth, Jan Zawallich
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE113 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-13374

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The generalized scalar auxiliary variable applied to the incompressible Boussinesq Equation Andreas Wagner, Barbara Wohlmuth, Jan Zawallich Abstract This paper introduces a second-order time discretization for solving the incompressible Boussinesq equation. It uses the generalized scalar auxiliary variable (GSAV) and a backward differentiation formula (BDF), based on a Taylor expansion around t n + k t^{n+k} for k ≥ 3 k\geq 3 . An exponential time integrator is used for the auxiliary variable to ensure stability independent of the time step size. We give rigorous asymptotic error estimates of the time-stepping scheme, thereby justifying its accuracy and stability. The scheme is reformulated into one amenable to a H 1 H^{1} -conforming finite element discretization. Finally, we validate our theoretical results with numerical experiments using a Taylor–Hood-based finite element discretization and show its applicability to large-scale 3-dimensional problems. ∎ MSC 46N40 65M12 65M15 1 Introduction The Boussinesq system zeytounian2003joseph for an incompressible fluid models the temperature evolution when gravity and buoyancy are the dominant forces on the fluid. For engineering applications, it is a vital component to describe cooling, ventilation, or the dispersion of heat de1983natural ; ward2000application ; BAKER1994261 ; lube2008stabilized . Aside from that, it is used in various scientific fields: It is integral for geophysical applications like earth-mantle convection, though here, the infinite Prandtl number approximation leads to an omission of the inertia terms in the fluid equation gassmoller2020formulations . Further, it is also applied to describe pollutant dispersion in environmental engineering MADALOZZO20145883 or to model the convection in stars spiegel1971convection in astrophysics. All in all, the applications are various, and hence, finding efficient discretizations for the Boussinesq system is of interest. Formally, the system can be stated in a non-dimensionalized form as follows: For the temperature \theta\colon[0,T]\times\Omega\to{\mathbb{R}} , the velocity {{\bf u}}\colon[0,T]\times\Omega\to{\mathbb{R}}^{d} , and the pressure p\colon[0,T]\times\Omega\to{\mathbb{R}} on the bounded domain \Omega\subset{\mathbb{R}}^{d} with time interval [0,T] , the incompressible Boussinesq system zeytounian2003joseph is given by \displaystyle\partial_{t}\theta+({{\bf u}}\cdot\nabla)\theta-\frac{1}{\textmd{Re Pr}}\Delta\theta= \displaystyle\,0 \displaystyle\textmd{ in }[0,T]\times\Omega, (1) \displaystyle\partial_{t}{{\bf u}}+({{\bf u}}\cdot\nabla){{\bf u}}-\frac{1}{\textmd{Re}}\Delta{{\bf u}}+\nabla p= \displaystyle\,\textmd{Ri }\theta{\bf e}_{g} \displaystyle\textmd{ in }[0,T]\times\Omega, (2) \displaystyle\nabla\cdot{{\bf u}}= \displaystyle\,0 \displaystyle\textmd{ in }[0,T]\times\Omega, (3) with the Reynolds number Re , Prandtl number Pr , and Richardson number Ri as material parameters, and where {\bf e}_{g} is the vector describing gravity. We use the boundary conditions {{\bf u}}=0 , and \partial_{n}\theta=0 on [0,T]\times\partial\Omega , and the initial conditions {{\bf u}}(0,\cdot)={{\bf u}}^{0} , \theta(0,\cdot)=\theta^{0} , for the purely spatial fields \theta^{0} , and {{\bf u}}^{0} . The model generally neglects temperature-induced density changes and only incorporates buoyancy effects due to gravity. Over the years, several time-stepping schemes have been proposed: In davis2002operator , a multiple-step time integrator is used for the fluid equation, while a backward Euler and a Crank-Nicolson scheme are applied to the energy equation. Both, fluid and energy equations use semi-implicit advective parts. A first-order convergent consistent projection scheme was proposed in yang2023error , including stability proofs and an error analysis incorporating the spatial discretization. In jiang2019pressure , a similar analysis is carried out for an ensemble scheme, including a particle mean velocity. Moreover, additional first and second-order IMEX schemes with stability proofs have been introduced in chen2023unconditional . A fully implicit scheme, including a nonlinear approximation, strategy has been proposed in damanik2009monolithic . In ding2024optimal , a second-order scheme including temperature-dependent viscosity and heat conduction coefficient has been introduced. In takhirov2021direction , the Boussinesq Equation on the spherical shell is solved with a direction splitting scheme using a domain-decomposition approach. Similarly, paszynski2020massively uses a splitting scheme and presents a massively parallel simulation for atmospheric flows. A mostly explicit method with an implicitly treated pressure term is presented in a high-performance context in castillo2024explicit with an analysis for the fluid-dynamic parts in kaya2024error . In pan2022monolithic , a staggered scheme for the momentum and energy equations is used, also considering additional temperature-related changes in the density and viscosity. A scheme with proofs for nonsmooth data was presented in zhang2018decoupled . Further, various schemes have been introduced when just the steady state is of interest hou2022decoupled ; hawkins2024analysis ; pollock2021acceleration . More recently, a scalar auxiliary variable (SAV) approach shen2018scalar ; shen2018convergence and more specifically, its extension to multiple scalar auxiliary variables (MSAV) cheng2018multiple , was applied to a Crank-Nicolson leap-frog scheme and a BDF2 scheme in jiang2023unconditionally in combination with stability terms for the spatial discretization. The unconditional stability was proven, and the second-order convergence was shown numerically. Though the scheme succeeds in linearizing the Boussinesq system, it still requires inverting two saddle point problems for the velocity and pressure, which typically requires advanced iterative linear solvers. In zhang2024error , a first-order SAV scheme based on a projection method is derived, including a stability and error analysis with remarks on how to generalize it to a second-order scheme. Recently, building on the SAV approach, a generalized scalar auxiliary variable (GSAV) approach huang2022new has been introduced for dissipative systems and successfully applied to the Navier–Stokes equation, obtaining a first-order in time consistent splitting scheme li2023error and a second-order scheme huang2023stability based on a BDF scheme using a time-discretization at t^{n+5} . These schemes require the inversion of two generalized Poisson equations per time step for the velocity and pressure, which is significantly easier to achieve, and computationally less expensive. Hence, in this paper, we build on the scheme introduced in huang2023stability to achieve a more efficient discretization for the Boussinesq equation. We thereby introduce a new auxiliary variable for stability and include temperature dependence in our error analysis. In addition, we justify the approach’s applicability for general Taylor expansions around t^{n+k} for k\geq 3 , instead of limiting it to k=5 . In the following, we want to present our time discretization for the Boussinesq equation, heavily relying on the analysis and time-stepping scheme of the incompressible Navier–Stokes in huang2023stability . We extend this scheme in Sec. 2 to the incompressible Boussinesq equation by including a bidirectional coupling to a temperature discretization. We will introduce two equivalent schemes, one having higher regularity requirements for the discretization inspired by huang2023stability , while for the other scheme, the ideas of shen2007error are used to rewrite our scheme in a feasible way for a finite element discretization. Our main theoretical result, an a priori error estimate for the time-discrete algorithm for d=2 , is introduced in Sec. 3 , with a rigorous proof in Sec. 4 . Finally, in Sec. 5 , we show numerical experiments validating our theoretical results, as well as 2D and 3D results for the Marsigli flow. 2 Time discretization and weak formulation We will consider a slightly more general problem than the instationary Boussinesq system. This problem is given by \displaystyle\partial_{t}\theta+({{\bf u}}\cdot\nabla)\theta-\kappa\Delta\theta= \displaystyle\,g \displaystyle\textmd{ in }[0,T]\times\Omega, (4) \displaystyle\partial_{t}{{\bf u}}+({{\bf u}}\cdot\nabla){{\bf u}}-\nu\Delta{{\bf u}}+\nabla p= \displaystyle\,{{\bf f}}(\theta) \displaystyle\textmd{ in }[0,T]\times\Omega, (5) with constraint \nabla\cdot{{\bf u}}=\,0\textmd{ in }[0,T]\times\Omega , and boundary conditions \theta=0,\textmd{ and }{{\bf u}}=0\textmd{ on }[0,T]\times\partial\Omega , with the right-hand side {{\bf f}}(t,\theta(t))={{\bf f}}_{1}(t)+{{\bf f}}_{2}(\theta(t)) . We assume that {{\bf f}}_{1} is uniformly bounded in time by the constant C_{{{\bf f}}_{1}} , and {{\bf f}}_{2} Lipschitz continuous, i.e. |{{\bf f}}_{2}(\theta)-{{\bf f}}_{2}(\tilde{\theta})|\leq\alpha|\theta-\tilde{\theta}| , and {{\bf f}}_{2}(0)=\mathbf{0} . Furthermore, g is uniformly bounded in time by C_{g} . Also, we assume appropriate initial conditions at t=0 for \theta ,

Read the complete text