Pattern formation in a Swift-Hohenberg equation with spatially periodic coefficients

Jolien Kamphuis, Martina Chirilus-Bruckner · 2025

Read Pattern formation in a Swift-Hohenberg equation with spatially periodic coefficients by Jolien Kamphuis, Martina Chirilus-Bruckner (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-2506-22211.txt.

Introduction

Jolien Kamphuis and Martina Chirilus-Bruckner study pattern growth in a classic model equation in 2025. The Swift-Hohenberg equation is a standard test case for how stripes and spots form. Here its terms vary in space in a repeating way, and the terms are large, not small.

The authors find a threshold where patterns arise as Bloch waves shaped by a second, simpler equation. They split the work into cases where the forcing rhythm matches the natural rhythm and cases where it does not. Working in Bloch space reveals richer forms. The frame should carry over to land plant models where ground shape varies.

Catalog record

CREATORJolien Kamphuis, Martina Chirilus-Bruckner
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE79 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-22211

Browse the full library · About the Tower of Babel

Full text

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

Pattern formation in a Swift-Hohenberg equation with spatially periodic coefficients Jolien Kamphuis, Martina Chirilus-Bruckner Abstract We study the Swift–Hohenberg equation – a paradigm model for pattern formation – with “large” spatially periodic coefficients and find a Turing bifurcation that generates patterns whose leading‑order form is a Bloch wave modulated by solutions of a Ginzburg–Landau type equation. Since the interplay between forcing wavenumber and intrinsic wavenumber crucially shapes the spectrum and emerging patterns, we distinguish between resonant and non‑resonant regimes. Extending earlier work that assumed asymptotically small coefficients, we tackle the more involved onset analysis produced by 𝒪 ⁡ ( 1 ) \mathcal{O}(1) forcing and work directly in Bloch space, where the richer structure of the bifurcating solutions becomes apparent. This abstract framework is readily transferable to more complex systems, such as reaction–diffusion equations arising as dryland vegetation models, where topography induces spatial heterogeneity. 1 Introduction Consider the Swift-Hohenberg equation \partial_{t}u=\left[-(\partial_{x}^{2}+k_{0}^{2})^{2}+p(x)\right]u-\rho(x)u^{3}\,, (1.1) for the unknown u=u(x,t)\in\mathbb{R} depending on x\in\mathbb{R},t\geq 0, and given k_{0}>0 and periodic coefficients \displaystyle p\left(x+\frac{2\pi}{k_{f}}\right)=p(x)\,,\quad\rho\left(x+\frac{2\pi}{k_{f}}\right)=\rho(x)\,,\quad x\in\mathbb{R}\,, (1.2) with period \frac{2\pi}{k_{f}} for some frequency k_{f}>0 . The classically studied Swift-Hohenberg equation (see [ 1 ] ) can be recovered by setting p(x)=\varepsilon^{2}r for some 0<\varepsilon\ll 1,r\in\mathbb{R},r=\mathcal{O}(1) , and \rho(x)=1 , x\in\mathbb{R} . For this choice the constant background solution u_{*}=0 undergoes a Turing bifurcation at r=0 giving rise to stable spatially periodic solutions around the wavenumber k_{0}>0 for r>0 . These are often referred to as patterns , since they are also directly linked to patterns in the two-dimensional case [ 2 ] . The dynamics of the amplitude of Turing patterns is known to be governed by the Ginzburg-Landau (GL) equation (see, e.g. [ 3 ] ) \partial_{T}A=rA+4k_{0}^{2}\partial_{X}^{2}A-3|A|^{2}A, (1.3) via the relation to the Swift-Hohenberg given by u(x,t)=u_{*}+\varepsilon A(X,T)e^{ik_{0}x}+\rm{c.c.}+\rm{h.o.t.}, (1.4) where X:=\varepsilon x , T:=\varepsilon^{2}t and \rm{h.o.t.} are higher order terms w.r.t. \varepsilon . Main objective. The present work studies pattern formation for the case of the Swift-Hohenberg equation with multiplicative spatially periodic forcing as given in ( 1.1 ) by rigorous derivation and justification of the Ginzburg-Landau equation. The evolution of patterns in systems with a slightly varying geometry has been previously studied by deriving the form of a Ginzburg-Landau equation for a class of reaction-diffusion equations with quadratic nonlinearity and small varying terrain, written as \nu p(k_{f}x) , with \nu=\mathcal{O}(\varepsilon^{2}) , k_{f}=\varepsilon\sigma_{0} , and \sigma_{0}=\mathcal{O}(1) in [ 4 ] , and with \nu=\mathcal{O}(\varepsilon) and k_{f}=\mathcal{O}(1) in [ 5 ] . These works were motivated by generalizing sedimentation in straight rivers, treated in [ 6 ] , to slightly meandering rivers. Ginzburg-Landau equations have also been derived for the Swift-Hohenberg model with small spatial forcing p(x)=\varepsilon^{2}r+\gamma\cos(k_{f}x) in [ 7 , 8 , 9 , 10 , 11 , 12 ] , where the forcing strength is taken of order \gamma=\mathcal{O}(\varepsilon^{k}) for k=1,2 . Furthermore, [ 13 , 14 ] considers a spatial forcing in the Lengyel-Epstein model. We will elaborate more on related work in Section 4 . Main novelty. Unlike previous studies, we allow forcing of order one, p(x)=\mathcal{O}(1) , whose amplitude and wavelength are neither small nor suited to classical averaging techniques. This regime introduces subtler bifurcation analysis due to the necessity of Bloch analysis at bifurcation onset, but it better reflects many applications. In particular, in the context of vegetation patterns, an application that partially motivated our work, p and \rho capture topographic effects. The abstract framework used here is readily transferable to more complicated settings relevant to applications and, in particular, lay groundwork for studying more complex PDEs such as ecologically motivated reaction–diffusion systems (cf. e.g. [ 15 ] ) and delivers the p(x)=\mathcal{O}(1) result that was not treated in [ 9 ] . Moreover, we give a validity proof and error estimates. We focus on the interaction between the system’s intrinsic wavenumber k_{0} and the forcing wavenumber k_{f} of the periodic coefficients p and \rho . The interplay between forcing strength, wavenumber k_{f} and intrinsic wavenumber k_{0} will give rise to resonant and non‑resonant parameter regimes. 1.1 Linear analysis In the classic case p(x)=\varepsilon^{2}r for some 0<\varepsilon\ll 1,r\in\mathbb{R},r=\mathcal{O}(1) and \rho(x)=1,x\in\mathbb{R} , a Turing instability is easily anticipated from the fact that all spatially bounded solutions of the linearized Swift-Hohenberg equation can be built from v(x,t)=e^{ikx}e^{\lambda(k)t} , \lambda(k)=-(k^{2}-k_{0}^{2})^{2}+\varepsilon^{2}r , where we depicted the so-called dispersion relation (k,\lambda(k)) where \lambda(k)=\lambda(k;k_{0},r) in Figure 1 . This shows that u_{*}=0 is linearly stable for r<0 and linearly unstable for r>0 . Upon adding nonlinear terms, this instability creates stable periodic solutions with wavenumber close to k_{0} . Note that, upon setting p(x)=\varepsilon^{2}r+p_{*}(x) with p_{*}(x)=\mathcal{O}(\varepsilon^{m}),m>0, and k_{f} non-resonant, the linearized problem at onset is to leading order in \varepsilon still the same as above, which makes the derivation of the GL equation similar to the constant-coefficient case. (a) Spectrum before onset (b) Spectrum at onset (c) Spectrum after onset Figure 1 : Spectrum of the linear operator of the constant coefficient Swift-Hohenberg system around u_{*}=0 . Before onset, r=-1 (left), at onset, r=0 (middle) and after bifurcation, r=1 (right). In stark contrast to this, the case of p(x)=\varepsilon^{2}r+p_{*}(x) with p_{*}(x)=\mathcal{O}(1) is more complicated, since spatially bounded solutions of the linearized PDE involving \displaystyle\mathcal{L}_{p}:=-\left(\partial_{x}^{2}-k_{0}^{2}\right)^{2}+p(x) (1.5) are built from Floquet-Bloch solutions (cf. [ 16 ] ) \displaystyle v(x,t)=e^{ilx}\psi_{n}(\ell,x)e^{\lambda_{n}(\ell)t}\,,\quad n\in\mathbb{N}\,, (1.6) with some function \psi_{n}\left(\ell,x+\frac{2\pi}{k_{f}}\right)=\psi_{n}(\ell,x) that has the same period as p,\rho and so-called band structure \displaystyle(\ell,\lambda_{n}(\ell))\,,\quad n\in\mathbb{N},-k_{f}/2\leq\ell<k_{f}/2\,, (1.7) with \lambda_{n}(\ell)=\lambda_{n}(\ell;k_{0},p) which, in general, cannot be computed explicitly (see Section 5.2 and [ 17 ] for an exception where it can be specified in terms of elementary functions) making numerical explorations indispensable. While the occurrence of Floquet-Bloch solutions is fairly generic and is backed by abstract theory, them having special features as need for a Turing instability is less explored analytically. We refrain from giving a full theoretical characterization of classes of forcing functions p (leaving it to future work) and rather work by assumptions, which also have the benefit of preparing further generalizations of the result. Hence, we will now collect a number of conditions on p that will give rise to a band structure that causes a Turing instability similar to the classic constant-coefficient SH (see Figure 2 for a cartoon). (a) Spectrum before onset (b) Spectrum at onset (c) Spectrum after onset Figure 2 : Cartoon of a band structure (\ell,\lambda_{n}(\ell)) , satisfying the Turing instability Assumption 1.1 : r<0,r=0,r>0 . 1.2 Main result Before stating the main result we compile a list of assumptions that it relies on. Assumption 1.1 (Turing instability) . Consider the linear operator \mathcal{L}_{p} in ( 1.5 ). We assume that \displaystyle p(x)=p_{*}(x)+\varepsilon^{2}r\,,\quad r\in\mathbb{R}\,, (1.8) for 0<\varepsilon\ll 1 and some periodic function p_{*}(x)=\mathcal{O}(1) is chosen such that the band structure ( 1.7 ) is given by (\ell,\lambda_{n}(\ell)) with \lambda_{n}(\ell)=\lambda_{n}(\ell;k_{0},p_{*},\varepsilon,r) and the following holds true for the Floquet-Bloch solutions from ( 1.6 ). (a) (ONB) The sequence of functions \{\psi_{n}(\ell,\cdot)\}_{n\in\mathbb{N}} forms an orthonormal basis of L^{2}\left(\left[0,\frac{2\pi}{k_{f}}\right]\right) for every -k_{f}/2\leq\ell<k_{f}/2 . (b) (Ordering of band structure) We have \lambda_{n}(\ell)\geq\lambda_{n+1}(\ell),n\in\mathbb{N},\lambda_{n}(\ell)\rightarrow-\infty,n\rightarrow\infty for every -k_{f}/2\leq\ell<k_{f}/2 . (c) (Regularity properties of spectral data) The \psi_{n},\lambda_{n} are at least three times differentiable w.r.t. \ell , with \partial_{\ell}^{k}\psi_{n}(\ell,x)\in L^{2}([0,2\pi/k_{f}]) for 0\leq k\leq 3 and satisfy \lambda_{n}(-\ell)=\lambda_{n}(\ell) , \psi_{n}(-\ell,x)=\overline{\psi_{n}(\ell,x)} for every -k_{f}/2\leq\ell<k_{f}/2 . (d) (Bifurcation w.r.t. \bm{r} ) The function p_{*}

Read the complete text