Resonances and computations

Yvain Bruned, Frédéric Rousset, Katharina Schratz · 2025

Read Resonances and computations by Yvain Bruned, Frédéric Rousset, Katharina Schratz (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-2504-19647.txt.

Introduction

Yvain Bruned and co-authors wrote a 2025 review of a new way to compute wave-like equations with rough data. Classic schemes often fail when the solution swings fast, which leads to blow-ups and loss of accuracy. Wave type equations are hard in part because they do not smooth rough starts.

The new approach studies the swings through resonance, a check of how waves interact. It pairs that check with a decorated tree scheme drawn from work on singular random equations. Trees keep track of each interaction step in a clean form. The review lays out the method and its roots for other scholars. The authors hope to see the scheme used for more wave equations and past them.

Catalog record

CREATORYvain Bruned, Frédéric Rousset, Katharina Schratz
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE155 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-19647

Browse the full library · About the Tower of Babel

Full text

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

Resonances and computations Yvain Bruned, Frédéric Rousset, Katharina Schratz Abstract The computation of time dynamics arising in nonlinear time-dependent partial differential equations is an ongoing challenge in numerical analysis, especially once roughness comes into play. Classical numerical schemes in general fail to resolve the oscillatory behaviour in the solution which leads to numerical instabilities and loss of convergence. Dispersive equations, e.g., nonlinear Schrödinger, Korteweg–de Vries and wave equations, thereby pose in particular a big problem as in contrast to the parabolic setting, no strong smoothing can be expected, i.e., if the initial data is rough, the solution stays rough which makes their approximation a delicate task. In this review we give an overview on a new numerical ansatz which aims to tackle the time dynamics of nonlinear dispersive partial differential equations even for very rough data. This is achieved by a resonance analysis and decorated tree formalism that draws its inpiration from the combinatorics used in the theory of regularity structures for solving singular SPDEs. One can hope to see this formalism applied in other contexts for dispersive PDEs and beyond. 1 Introduction The numerical computation of nonlinear partial differential equations (PDEs) for smooth solutions is nowadays in large parts well understood. A large toolbox of techniques to approximate effectively their time dynamics was established, involving splitting methods (with the intention of splitting the full equation into a series of simpler subproblems), exponential integrators (based on Duhamel’s formula) and many more approaches. For an extensive overview of classical methods in context of dispersive PDEs we refer to [ 29 , 31 , 37 , 37 , 42 , 68 ] and the references therein. While classical techniques in many situations allow an efficient and well-suited numerical approximation, they in general fail to reproduce the behaviour of solutions once roughness comes into play. This is due to the fact that in general classical integration methods require a lot of regularity on the solutions in order to converge. If the solution is not regular enough, well-known theoretical error estimates break down and almost nothing is known on how to overcome this. Break down of convergence for rough data is not only a theoretical artefact, but also severely observed in numerical simulations. If the solution does not meet the required smoothness assumptions, the numerical solution becomes unstable and eventually blows up. More details on the struggle of classical methods for rough data are given in Section 2 . The main question we want to address in this review is the following: How can we actually reproduce rough time dynamics in nonlinear PDEs an overcome classical smoothness assumptions? To answer this question we will present a novel integration technique, so-called resonances as a computational tool , with the aim to reduce as far as possible the high regularity assumptions needed in classical schemes. The novel idea of this approach lies in building numerical schemes on the underlying structure of resonances of the PDE, instead of discretising the PDE itself. This is achieved by looking at iterations of Duhamel’s formula and filtering out the dominant parts in the oscillations. To control the higher order oscillations, we use a decorated tree formalism with the frequencies as decorations. This will allow us to construct resonance based schemes up to arbitrary order. Details are given in Section 3 where we present the main ideas behind the derivation of low regularity schemes for dispersive equations. We also present some extensions that allow us to go deeper, e.g., preserving essential symmetries in the equation. 2 Classical methods and their problem for rough data To present the main challenges of classical methods for rough data, let us consider as a model problem the periodic Korteweg–de Vries equation \partial_{t}u(t,x)+\partial_{x}^{3}u(t,x)=\frac{1}{2}\partial_{x}u^{2}(t,x)\qquad(t,x)\in\mathbb{R}\times\mathbb{T} (1) with rough initial data u(0,x)=u_{0}(x)\in H^{\sigma}(\mathbb{T}), (2) where we want to choose \sigma>0 as small as possible (we will see later how far we can actually push down the Sobolev index \sigma ). Here, H^{\sigma}(\mathbb{T}) denotes the classical Sobolev space on torus \mathbb{T} with regularity \sigma\geq 0 . Before we present the main idea of resonances as a computational tool – a new method which ideally allows us to choose \sigma in ( 2 ) arbitrarily close to zero – let us first give a brief overview of classical numerical methods to solve the nonlinear PDE ( 1 ). This will allow us to better understand the regularity assumptions of classical approaches and get essential insight on how to overcome this. 2.1 Splitting methods One of the most famous numerical methods to approximate the time dynamics of KdV equation ( 1 ) and in general nonlinear dispersive models are splitting methods (see, e.g., [ 31 , 37 , 39 , 44 , 46 , 55 , 57 , 48 ] ). The main idea thereby lies in splitting the full equation into a series of simpler subproblems. One then solves these subproblems (either exactly or with a numerical scheme) and composes the sub-flows to obtain an approximation to the flow of the original – full – equation. Looking at the KdV equation ( 1 ) we face two main challenges numerically: The differential operator \partial_{x}^{3} and the non-linear Burger’s term \frac{1}{2}\partial_{x}u^{2} . The idea of splitting is now the following: Instead of solving the full problem ( 1 ) we consider its linear and nonlinear part separetely, i.e., \displaystyle\text{ (L) }\quad\partial_{t}u(t,x)=-\partial_{x}^{3}u(t,x)\quad\text{and}\quad\text{ (N) }\quad\partial_{t}u(t,x)=\frac{1}{2}\partial_{x}u^{2}(t,x). (3) The main advantage thereby lies in the fact that the time dynamics in both subproblems can be solved exactly: (L) in Fourier space (in case of a spatial discretisation we can couple this with for instance a Fourier pseudo spectral method) and (N) with the method of characteristics. Remark 1 In the end we are of course interested in a full discretisation as we have a PDE both in time and in space. This will introduce an additional spatial error. In case of splitting for KdV we will have a spectral error for the linear problem (L) as well as an interpolation error for the nonlinear problem (N) as we do not necessarily end up on the grid points any more when using the method of characteristics. The latter requires an interpolation, e.g., Lagrange interpolation, or the use of WENO schemes, e.g., upwind scheme, see for instance [ 1 , 53 ] . In this review we only focus on the time dynamics and hence neglect the additional error introduced by the spatial discretisation of PDEs. For fully discrete resonance based schemes in context of Navier–Stokes equation and Schrödinger equations we refer for instance to [ 51 , 47 ] . A simple composition of the kinetic flow \phi_{L}^{t} and nonlinear flow \phi_{N}^{t} leads at first-order to the so-called first order Lie splitting method (with time step size \tau ) \phi_{L}^{\tau}\circ\phi_{N}^{\tau} (4) and its symetrised version the classical second-order Strang splitting \phi_{L}^{\tau/2}\circ\phi_{N}^{\tau}\circ\phi_{L}^{\tau/2}. (5) We refer to an overview of splitting methods to [ 37 , 44 , 57 , 48 ] . Here we are in interested in the local error structure the splitting approach introduces for KdV, as this will allow us to answer the question what kind of regularity assumptions this imposes on the solution and more importantly, how small can we actually choose the regularity

Read the complete text