An Augmented Lagrangian Preconditioner for Navier–Stokes Equations with Runge–Kutta in Time

Santolo Leveque, Yunhui He, Maxim Olshanskii · 2025

Read An Augmented Lagrangian Preconditioner for Navier–Stokes Equations with Runge–Kutta in Time by Santolo Leveque, Yunhui He, Maxim Olshanskii (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-2506-04451.txt.

Introduction

Santolo Leveque, Yunhui He, and Maxim Olshanskii make a hard fluid computation cheaper. They time-step the incompressible Navier-Stokes equations with Runge-Kutta methods. Each stage yields a large nonlinear system. Each Newton iteration inside it faces a sprawling saddle-point linear problem, and how well those systems are solved decides if the whole scheme is practical.

Their augmented Lagrangian preconditioner works inside a flexible iterative solver. It can be applied inexactly through multigrid. Numerical experiments vary viscosity, mesh size, time step, and stage count to stress the design. The preconditioner stayed robust and efficient across those changes. The work targets engineers and numerical analysts who simulate unsteady flows at scale.

Catalog record

CREATORSantolo Leveque, Yunhui He, Maxim Olshanskii
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-04451

Browse the full library · About the Tower of Babel

Full text

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

An Augmented Lagrangian Preconditioner for Navier–Stokes Equations with Runge–Kutta in Time Santolo Leveque, Yunhui He, Maxim Olshanskii Abstract We consider a Runge–Kutta method for the numerical time integration of the nonstationary incompressible Navier–Stokes equations. This yields a sequence of nonlinear problems to be solved for the stages of the Runge–Kutta method. The resulting nonlinear system of differential equations is discretized using a finite element method. To compute a numerical approximation of the stages at each time step, we employ Newton’s method, which requires the solution of a large and sparse generalized saddle-point problem at each nonlinear iteration. We devise an augmented Lagrangian preconditioner within the flexible GMRES method for solving the Newton systems at each time step. The preconditioner can be applied inexactly with the help of a multigrid routine. We present numerical evidence of the robustness and efficiency of the proposed strategy for different values of the viscosity, mesh size, time step, and number of stages of the Runge–Kutta method. keywords Time-dependent problems, Parabolic PDEs, Preconditioning, Saddle-point systems AMS 65F08, 65F10, 65N22, 65L06 1 Introduction Accurately modeling the motion of Newtonian incompressible viscous fluids is a fundamental problem in computational science and engineering. The governing equations for such flows are the incompressible Navier–Stokes equations, whose numerical solution has been the focus of extensive research for decades. Many time-stepping schemes for the nonstationary Navier–Stokes equations require solving a nonlinear problem at each time step, typically via Picard or Newton iterations. These nonlinear solvers in turn generate a sequence of linear systems that must be solved efficiently at each iteration. Classical time integration schemes for the nonstationary incompressible Navier–Stokes equations include multistep methods such as backward Euler and the trapezoidal rule [ 48 , 53 ] , projection methods [ 20 ] , and implicit-explicit (IMEX) methods [ 32 ] , among others. There are many good reasons for using these methods. First, the block structure of the discretized system allows one to adapt linear solvers for the stationary Navier–Stokes problem to the nonstationary case. Furthermore, some of these methods are A-stable, so no Courant–Friedrichs–Lewy (CFL) condition must be satisfied, meaning that no restriction on the time step is required. These methods satisfy a proper energy balance that ensures energy stability. Despite the advantages described above, classical time integrators for the nonstationary incompressible Navier–Stokes equations are typically of order no greater than two. Specifically, for linear multistep methods, the second Dahlquist barrier states that an A-stable linear multistep method cannot have an order of convergence greater than two; see, for example, [ 28 , Theorem 6.6] . However, the work in [ 22 ] develops high-order (up to fifth-order) A-stable IMEX methods for the nonstationary incompressible Navier–Stokes equations under periodic boundary conditions. In contrast to linear multistep methods, it is possible to devise A-stable implicit Runge–Kutta methods of arbitrary order. Moreover, some implicit Runge–Kutta methods can be constructed to be L- or B-stable; see, for instance, [ 13 , 21 , 28 ] . This motivates us to consider Runge–Kutta methods for the nonstationary incompressible Navier–Stokes equations. Although there are several studies applying Runge–Kutta methods to the time integration of the Navier–Stokes equations, we are aware of only [ 2 ] that addresses the linear solver aspect of this approach. Specifically, the authors of [ 2 ] designed a monolithic multigrid method as the stage solver. The recent rediscovery of Runge–Kutta methods as time integrators for the numerical solution of PDEs has led to a plethora of linear solvers for the system of stages of the method. After the development of simple iterations such as the block-Jacobi [ 31 ] and block-Gauss–Seidel [ 47 ] preconditioners for the heat equation, more advanced techniques have been employed to find good approximations of the stage systems. For example, strategies based on special LU factorizations of the coefficient matrix [ 37 ] and diagonalizations of its lower triangular factor [ 3 ] have been proposed. Further, in [ 34 ] , the authors develop a stage-parallel implementation of the preconditioner derived in [ 3 ] . Regarding parallel implementations of Runge–Kutta integrators for PDEs, we mention [ 27 , 30 ] , where the authors propose parallel-in-time solvers for the all-at-once space-time discretization of the heat equation with Runge–Kutta in time. The solver in [ 30 ] is further adapted to handle the incompressible Stokes equations. Other examples of stage solvers include methods based on the eigenvalue decomposition of the inverse of the coefficient matrix [ 46 ] , the Schur decomposition of the coefficient matrix [ 45 ] , and monolithic multigrid solvers [ 2 , 15 , 25 ] . We also mention early works on monolithic multigrid methods for stage solvers by Vandewalle and co-authors [ 9 , 39 , 50 ] . The preconditioning strategy we adopt here is based on an augmented Lagrangian approach. Originally, the augmented Lagrangian preconditioner was introduced in [ 7 ] for solving the Picard linearization of the stationary incompressible Navier–Stokes equations. This preconditioning strategy was shown to be robust with respect to the mesh size and the viscosity of the fluid—features that are particularly important for solver performance. The key components of the preconditioner are a specialized multigrid routine used for the approximate inversion of the augmented momentum equation and a special approximation of the Schur complement, derived from the application of the Sherman–Morrison–Woodbury formula. In [ 8 ] , the authors proved the optimality of the preconditioner via a field-of-values analysis of the preconditioned Schur complement, showing that its eigenvalues are contained in a rectangle in the right half of the complex plane, with boundaries independent of the viscosity, provided the augmentation parameter is proportional to the inverse of the viscosity. However, as demonstrated by the numerical results in [ 7 ] , robust convergence is also achieved when the augmentation parameter is of order one, provided a suitable rescaling of the equations is applied. In [ 16 ] , the authors apply this augmented Lagrangian strategy to the Newton linearization of the 3D stationary incompressible Navier–Stokes equations. In this case as well, the numerical results confirm the effectiveness and robustness of the approach. Finally, we mention that in [ 29 ] , an augmented Lagrangian-based preconditioner is employed in an inexact Newton iteration for the stationary incompressible Navier–Stokes control problem. As we will show in the numerical results, the augmented Lagrangian-based preconditioner we adopt in this work is robust with respect to both the problem parameters and the choice of Runge–Kutta method, resulting in a linear solver that scales linearly with the problem size. This paper is organized as follows. In Section 2 , we introduce Runge–Kutta methods, which are then applied to the nonstationary incompressible Navier–Stokes equations in Section 3 . In Section 4 , we present the augmented Lagrangian preconditioner used to solve the stage systems of the Runge–Kutta method. Numerical evidence of the efficiency and robustness of the preconditioner is given in Section 5 . Finally, conclusions and directions for future work are discussed in Section 6 . 2 Runge–Kutta Methods We recall a variant of the Runge–Kutta method for an index 2 differential–algebraic equation (DAE) of the form \left\{\begin{array}[]{ll}\vskip 1.07639pt\dot{u}=\mathcal{F}(\hat{w},t)&\quad\mathrm{in}\;(0,T),\\ \vskip 1.07639pt\mathcal{G}(u,t)=0&\quad\mathrm{in}\;(0,T),\end{array}\right. (1) with \hat{w}=(u,p) and a suitable

Read the complete text