Stability, convergence, and geometric properties of second-order-in-time space–time discretizations for linear and semilinear wave equations
2026
Read Stability, convergence, and geometric properties of second-order-in-time space–time discretizations for linear and semilinear wave equations (2026) 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-2601-03160.txt.
Introduction
Matteo Ferrari, Ilaria Perugia, and Enrico Zampa revisit numerical schemes that solve wave equations by treating space and time together, with second-order accuracy in time. They establish precise equivalences between these space-time discretizations and first-order-in-time formulations. The schemes use continuous piecewise-polynomial trial functions in time. These are tested against discontinuous polynomials of one degree lower.
With that setup, one popular scheme turns out to match a method from the nineteen nineties, in the linear case, and its semilinear extension. It conserves energy at mesh nodes but is not symplectic. The authors then introduce two symplectic variants, built from Gauss-Legendre and Gauss-Lobatto quadrature in time. Each corresponds to a specific Runge-Kutta integrator. The connections clarify the geometric structure behind the methods.
Catalog record
| YEAR | 2026 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 89 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-03160
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Stability, convergence, and geometric properties of second-order-in-time space–time discretizations for linear and semilinear wave equations Abstract We revisit second-order-in-time space–time discretizations of the linear and semilinear wave equations by establishing precise equivalences with first-order-in-time formulations. Focusing on schemes using continuous piecewise-polynomial trial functions in time, we analyze their stability, convergence, and geometric properties. We consider first a weak space–time formulation with test functions projected onto discontinuous polynomials of one degree lower in time, showing that it is equivalent to the scheme proposed in [French, Peterson 1996] in the linear case, and extended in [Karakashian, Makridakis 2005] to the semilinear case. In particular, this equivalence shows that this method conserves energy at mesh nodes but is not symplectic. We then introduce two symplectic variants, obtained through Gauss–Legendre and Gauss–Lobatto quadratures in time, and show that they correspond to specific Runge–Kutta time integrators. These connections clarify the geometric structure of the space–time methods considered. mathx”17 1 Introduction We study a class of second-order-in-time space–time Galerkin methods for the linear and semilinear wave equations and establish novel equivalence results with first-order-in-time formulations. These equivalences enable the rigorous transfer of known properties from first-order to second-order schemes. Second-order-in-time space–time Galerkin schemes for the wave equation have recently attracted significant interest. A common construction begins by multiplying the wave equation by test functions of space and time that vanish at the final time, followed by integration by parts in both space and time. The initial condition for the solution is then imposed strongly, while the initial condition for its time derivative is enforced weakly. The resulting formulation is then discretized with continuous finite elements in both space and time. For the linear wave equation, it is well known that such a formulation may fail to be unconditionally stable when standard discrete time spaces are used; see [ 36 , Remark 4.9] for continuous piecewise linear polynomials, and [ 8 , Theorem 5.9] for maximal regularity splines. This instability arises from the lack of inf–sup stability in the energy norm [ 40 , Theorem 4.2.23] . To overcome this issue, various stabilization techniques have been proposed. In [ 35 ] , a stabilization technique was introduced for continuous piecewise linear polynomials in time, in which the test function in the spatial stiffness term is projected onto a space of piecewise constant functions in time. This approach was later extended and numerically validated for higher-order continuous piecewise polynomials in [ 41 ] , by projecting onto discontinuous piecewise polynomial of one degree lower. For continuous piecewise linear polynomials, the stabilization proposed in [ 35 ] is equivalent to adding a non-consistent penalty term. Along these lines, a stabilization method was proposed in [ 10 ] (and subsequently analyzed in [ 8 ] ) for maximal regularity splines in time. Another approach involves transforming the test functions using an operator: the modified Hilbert transform, which has yielded promising numerical results in [ 28 ] , or via Morawetz multiplier [ 3 ] . In this paper, we prove that a slight modification of the stabilized formulation of [ 41 ] is equivalent to the first-order-in-time discontinuous Galerkin–continuous Galerkin (DG–CG) method introduced in [ 1 , 11 ] for the linear wave equation. This implies that the second-order-in-time method possesses the same theoretical properties that have been established for the first-order-in-time method in [ 11 , 13 ] . Since the modification affects only the right-hand side term, the stability results also carry over to the original formulation of [ 41 ] . This equivalence generalizes the result of [ 42 ] , which showed that on tensor product meshes the lowest-order discretization in [ 41 ] coincides with the Crank–Nicolson scheme in time (equivalently, Newmark with parameters (\beta,\gamma)=(1/4,1/2) ). Up to the treatment of the right-hand side, this scheme is itself equivalent to the lowest-order DG–CG method. We then apply the same framework to the semilinear wave equation. In all terms of the resulting bilinear form, except for the temporal stiffness term, a projection onto the space of discontinuous polynomials of one degree lower is inserted. The resulting formulation is equivalent to the extension of the first-order-in-time DG–CG method analyzed in [ 21 ] , so that the stability and convergence established there carry over directly to the second-order-in-time scheme. While preserving the unconditional stability and convergences properties established in [ 11 , 19 , 13 ] , the second-order-in-time formulation reduces the number of unknowns by half. Moreover, although it uses trial and test functions that are continuous in time, it can still be implemented in a time-stepping fashion and efficient solvers are available, see [ 43 , §7] (and also [ 38 , 22 , 27 ] ). The methods discussed so far exactly conserve a discrete energy. In practice, however, nonlinear terms are integrated using quadrature, so energy conservation holds only up to quadrature error. Quadrature can also be exploited to preserve other geometric properties, such as symplecticity. Building on this, we propose two symplectic variants based on Gauss–Legendre and Gauss–Lobatto quadratures. On tensor product meshes, these correspond, respectively, to applying the Gauss–Legendre Runge–Kutta or the Lobatto IIIA/IIIB Runge–Kutta time discretizations to the first-order-in-time equivalent formulation. As a result, they are symplectic, with the Gauss–Legendre variant unconditionally stable and the Gauss–Lobatto variant only conditionally stable. We summarize the main properties of the considered methods in Table 1 . Method Stability Energy preservation Symplecticity Unstabilized \triangle \times \checkmark Stabilized \checkmark \checkmark (up to quadrature error) \times Gauss–Legendre \checkmark \times \checkmark Gauss–Lobatto \triangle \times \checkmark Table 1 : Properties of four second-order-in-time methods: “Unstabilized” is the basic method, “Stabilized” refers to the method with the projection onto discontinuous polynomials of one degree lower, and “Gauss–Legendre” and “Gauss–Lobatto” are the methods with the corresponding quadratures. \checkmark = yes, \times = no, \triangle = conditional. Outline. The paper is organized as follows. In Section 2 , we focus on the second-order-in-time space–time method for the linear wave equation and establish its equivalence with the first-order-in-time formulation. This yields an unconditionally stable scheme that can be efficiently implemented using under-integration in time. In Section 3 , we extend the same framework to the semilinear wave equation. Section 3.3 presents symplectic variants based on Gauss–Legendre and Gauss–Lobatto quadratures, which preserve key geometric properties. Numerical results reported in sections 2.5.1 and 3.2.1 validate, respectively, energy preservation in the linear case and the convergence rates for the sine-Gordon equation. 2 The linear case In this section, we consider a second-order-in-time space–time method for the linear wave equation, which is a slight modification of the stabilized approach from [ 41 ] . We prove an equivalence result between this formulation and the space–time Discontinuous–Continuous Galerkin (DG–CG) discretization of the problem in its first-order-in-time form, as introduced in [ 1 , 11 ] , allowing theoretical results established for the latter to be extended to the former. We also describe an efficient implementation of the second-order-in-time method that avoids solving local linear systems for computing the L^{2} projections required by the formulation. 2.1 The linear wave equation We consider the following initial-value problem with homogeneous Dirichlet boundary conditions \begin{cases}\partial_{t}^{2}U-\text{div}_{\boldsymbol{x}}(c^{2}\nabla_{\boldsymbol{x}}U)=F&\text{in~}Q_{T}:=\Omega\times(0,T),\\ U=0&\text{on~}\partial\Omega\times(0,T),\\ U(\cdot,0)=U_{0},\quad\partial_{t}U(\cdot,t)_{|_{t=0}}=V_{0}&\text{in~}\Omega,\end{cases} (1) where \Omega\subset\mathbb{R}^{d} ( d=1,2,3 ) is a polytopal Lipschitz domain, T>0 is a