1 Introduction

Our goal in this paper is to complete the program initiated in [28, 29] to analyze the impact of adaptive discretization schemes on scalar bistable reaction–diffusion PDEs of the form

$$\begin{aligned} u_t = u_{xx} + g(u). \end{aligned}$$
(1.1)

In particular, for any discretization distance \(h >0\) and any \(j \in {\mathbb {Z}}\), we write \(x_{jh}(t)\) for the time-dependent location of the relevant gridpoint and \(U_{jh}(t)\) for the associated approximation for \(u\big (x_{jh}(t), t\big )\). We then study the system

$$\begin{aligned} \begin{array}{lll} {\dot{U}}_{jh}(t) &{} = \Big [ \frac{U_{(j+1)h}(t) - U_{(j-1)h}(t)}{x_{(j+1)h}(t) - x_{(j-1)h}(t)} \Big ] {\dot{x}}_{jh}(t)\\ &{} \quad + \frac{2}{x_{(j+1)h}(t) - x_{(j-1)h}(t)} \Big [ \frac{U_{(j-1)h }(t) - U_{jh}(t)}{x_{jh}(t) - x_{(j-1)h}(t) } + \frac{U_{(j+1)h}(t) - U_{jh}(t)}{x_{(j+1)h}(t) - x_{jh}(t)} \Big ] + g\big ( U_{jh}(t) \big ),\\ \end{array} \end{aligned}$$
(1.2)

in which x(t) is defined implicitly by demanding that

$$\begin{aligned} \begin{array}{lcl} \big (x_{(j+1)h}(t) - x_{jh}(t) \big )^2 + \big ( U_{(j+1)h}(t) - U_{jh}(t) \big )^2 = h^2 \end{array} \end{aligned}$$
(1.3)

and imposing the boundary constraint

$$\begin{aligned} \lim _{j \rightarrow - \infty } [x_{jh}(t) - jh ] = 0. \end{aligned}$$
(1.4)

This means that mesh locations are determined dynamically in time so that there is equidistribution of a finite difference approximation of the arclength of the solution of (1.2). We show that this system has solutions of the form

$$\begin{aligned} U_{jh}(t) = \Phi ( x_{jh}(t) + ct ), \end{aligned}$$
(1.5)

which can be interpreted as travelling waves. For concreteness, we will use the cubic nonlinearity

$$\begin{aligned} g(u) = g_{\mathrm {cub}}(u;a) = u ( 1- u) (u - a), \qquad \qquad 0< a < 1 \end{aligned}$$
(1.6)

throughout this introduction to explain the main ideas.

Travelling waves The pair \((\Phi ,c)\) that we construct will be close to the travelling wave \((\Phi _*, c_*)\) for the PDE (1.1). Using (1.6), this pair must satisfy the travelling wave ODE

$$\begin{aligned} c\Phi _*' = \Phi _*'' + g_{\mathrm {cub}} (\Phi _* ; a), \qquad \qquad \Phi _*(-\infty ) = 0, \qquad \Phi _*(+\infty ) = 1. \end{aligned}$$
(1.7)

Such solutions provide a mechanism through which the fitter biological species (corresponding to the deepest well of the potential \(- \int g_{\mathrm {cub}}\)) can become dominant throughout a spatial domain. For this reason they are sometimes referred to as invasion waves.

It is well-known that these waves play an important role in the global dynamics of (1.1). For example, using the comparison principle one can show that these waves are nonlinearly stable under a large class of perturbations [15] and that they determine the spreading speed of localized structures [42]. In addition, they have been used extensively as building blocks to construct general time dependent solutions of reaction–diffusion systems. For example, planar versions of (1.1) support (sharp) travelling corners [5, 17], expansion waves [36], scattering waves [4] and modulated waves [10] that connect periodic travelling waves of nearby frequencies.

Uniform spatial discretizations In order to set the stage, let us return to the lattice differential equation (LDE)

$$\begin{aligned} {\dot{U}}_j(t) = \frac{1}{h^2}[U_{j-1}(t) + U_{j+1}(t) -2 U_j(t) ] + g_{\mathrm {cub}}\big (U_j(t);a \big ), \end{aligned}$$
(1.8)

which can be used to describe the uniformly discretized approximants \(U_{j}(t) \sim u(jh , t)\). Mathematically speaking, the transition from (1.1) to (1.8) breaks the continuous translational symmetry of the underlying space. Indeed, (1.8) merely admits the discrete group of symmetries \(j \mapsto j + k\) with \(k \in {\mathbb {Z}}\). As a consequence, travelling wave solutions

$$\begin{aligned} U_j(t) = \Phi (j h+ct) \end{aligned}$$
(1.9)

can no longer be seen as equilibria in an appropriate comoving frame. Instead, they must be treated as periodic solutions modulo the discrete shift symmetry discussed above. The resulting challenges occur frequently in similar discrete settings and general techniques have been developed to overcome them [3, 8, 16].

Direct substitution of (1.9) into (1.8) yields the travelling wave equation

$$\begin{aligned} c\Phi '(\xi ) = \frac{1}{h^2}[\Phi (\xi -h) + \Phi (\xi +h) -2\Phi (\xi )] +g_{\mathrm {cub}}(\Phi (\xi ) ; a ), \end{aligned}$$
(1.10)

to which we again append the boundary conditions

$$\begin{aligned} \Phi (-\infty ) = 0, \qquad \qquad \Phi (+ \infty ) = 1. \end{aligned}$$
(1.11)

Due to the presence of the shifted arguments such equations are known as functional differential equations of mixed type (MFDEs). Note that the unbounded second derivative operator in (1.7) has been replaced by a bounded second-difference operator. In addition, the transition \(c \rightarrow 0\) is now singular since it changes the structure of the equation. As a consequence, there is a fundamental difference between standing and moving wave solutions to (1.8).

In the anti-continuum regime \(h \gg 1\), the second-difference operator can be treated as a small perturbation to the remaining ODE. An elegant construction pioneered by Keener [31] allows one to construct standing waves for \(a \ne \frac{1}{2}\) that satisfy the boundary conditions (1.11) and block the two stable background states \(\Phi \equiv 0\) and \( \Phi \equiv 1\) from invading the domain. In particular, the shape of the potential \(-\int g_{\mathrm {cub}}\) no longer determines the sign of the wave speed. This phenomenon is often referred to as pinning or propagation failure and has attracted a considerable amount of attention [1, 9, 11, 12, 19, 24].

In the intermediate \(h \sim 1\) regime the shifted terms cannot be handled so easily and one needs to understand the full MFDE. Such equations are ill-posed as initial value problems and hence must be handled delicately. Several important tools have been developed to accomplish this, such as Fredholm theory [32] and exponential dichotomies [18, 34, 37, 38]; see [23] for a detailed overview.

Using a global homotopy argument together with the comparison principle, Mallet-Paret [33] constructed a branch of solutions \(\big (\Phi (a), c(a)\big )\) to (1.10) with (1.11), in which c(a) is unique and \(\Phi (a)\) is unique up to translation when \(c(a) \ne 0\). For the uniform spatial discretization of the FitzHugh-Nagumo PDE [28,  Eq. (1.3)], a generalization of Lin’s method can be used to establish a version of the exchange Lemma for MFDEs and construct stable travelling pulses [25, 26]. Further results in this area can be found in [6, 7, 31, 33, 43] and the survey [23].

Finally, in the continuum regime \(0 < h \ll 1\) it is natural to treat the second-difference operator in (1.10) as a perturbation to the second-derivative in (1.7). Johann [30] developed a version of the implicit function theorem that can handle such singular perturbations in some settings. In addition, the spectral convergence approach developed by Bates and his coauthors in [2] can be used to study the impact of this transition directly on the linear operators involved. We used the main ideas behind this technique in [29] to provide the key linear result that underpins the nonlinear analysis in this paper.

Uniform spatial-temporal discretizations The so-called backward differentiation formula (BDF) are a family of six schemes that can be used for discretizing the temporal derivative in (1.8). These are well-known multistep methods that are appropriate for parabolic PDEs due to their numerical stability properties. As an illustration, we note that the two lowest order schemes prescribe the substitutions

$$\begin{aligned} \begin{array}{lcl} {\dot{U}}_j(t) &{} \mapsto &{} \frac{1}{\Delta t} \big [ U_{ j}\big (n \Delta t \big ) - U_{ j}\big ( (n-1)\Delta t \big ) \big ], \\ {\dot{U}}_j(t) &{} \mapsto &{} \frac{1}{2 \Delta t} \big [ 3 U_{ j}\big (n \Delta t \big ) - 4U_{ j}\big ( (n-1)\Delta t \big ) + U_{ j}\big ( (n-2)\Delta t \big ) \big ], \end{array} \end{aligned}$$
(1.12)

in which \(\Delta t > 0\) denotes the timestep and \(t = n \Delta t\). The first scheme is also known as the backward Euler method and has the advantage that it preserves the comparison principle, unlike the five other members of the family.

In [27] we constructed fully discretized travelling wave solutions

$$\begin{aligned} U_j(n \Delta t) = \Phi (j + n c \Delta t), \qquad \Phi (-\infty ) =0, \qquad \Phi (+\infty ) = 1 \end{aligned}$$
(1.13)

for the coupled map lattices arising from these discretization schemes. The relevant travelling wave equations for the two lowest order schemes can be obtained by making the replacements

$$\begin{aligned} \begin{array}{lcl} c \Phi '(\xi ) &{} \mapsto &{} \frac{1}{\Delta t} \big [ \Phi (\xi ) - \Phi (\xi - c \Delta t) \big ], \\ c \Phi '(\xi ) &{} \mapsto &{} \frac{1}{2\Delta t} \big [ 3 \Phi (\xi ) - 4\Phi (\xi - c \Delta t) + \Phi (\xi - 2 c \Delta t) \big ] \end{array} \end{aligned}$$
(1.14)

in the MFDE (1.10). In the first case we leveraged the comparison principle to obtain global results. We established that the c(a) relation can become multi-valued, which clearly distinguishes the fully-discrete regime from its spatially-discrete counterpart. The same behaviour occurs for the five other BDF methods, but here we only have results for small \(\Delta t > 0\). We remark that related phenomena have been observed in monostable KPP systems [35] in the presence of inhomogeneities.

These non-uniqueness results should be seen as part of the program that was initiated in [12,13,14] to study the impact of temporal and full discretization schemes on various reaction–diffusion systems. Indeed, these papers studied versions of (1.1) with various smooth and piecewise linear bistable nonlinearities. The authors used adhoc techniques to obtain rigorous, formal and first order information concerning the change in the dynamics of traveling wave solutions. In addition, in [8] the authors considered the forward-Euler scheme and used Poincaré return-maps and topological arguments to obtain the existence of fully-discretized waves.

Computational frame In [28] we showed that the dynamics of the coupled system (1.2)–(1.4) can be reduced to an equivalent system of the form

$$\begin{aligned} {\dot{U}}_{kh} = {\mathcal {G}} \Big ( \{U_{jh}\}_{j \le k +1 } \Big ), \end{aligned}$$
(1.15)

in which \({\mathcal {G}}\) is a (convoluted) nonlinear expression that we describe explicitly in Sect. 2. In order to appreciate this equation, it is insightful to transform (1.1) into a new coordinate system \((\theta , t)\) by demanding \(\theta _x = \sqrt{1 + u_x^2}\). Indeed, in these new arclength coordinates the transformed functions

$$\begin{aligned} w(\theta , t) = u\big ( x(\theta , t), t\big ), \qquad \qquad \gamma (\theta , t) = \sqrt{1 - w_{\theta }(\theta ,t)^2} \end{aligned}$$
(1.16)

can be shown [28,  §1] to satisfy the nonlocal PDE

$$\begin{aligned} w_t = \gamma ^{-2} w_{\theta \theta } + \gamma ^2 g(w) + w_\theta \int _- \Big (\gamma ^{-4} w_{\theta \theta } + g(w) \Big ) w_{\theta \theta }, \end{aligned}$$
(1.17)

in which we use the notation \([\int _- f](\theta ) = \int _{-\infty }^\tau f(\theta ') \, d \theta '\). This coincides with the system that arises by taking the formal \(h \downarrow 0\) limit in (1.15).

In [29] we constructed a solution to (1.17) by stretching the PDE waveprofile \(\Phi _*\) into its arclength parametrized form \(\Psi _*\) and writing

$$\begin{aligned} w(\theta , t)= \Psi _*(\theta + c t). \end{aligned}$$
(1.18)

Motivated by this observation, the main goal of this paper is to find solutions to (1.15) of the form

$$\begin{aligned} U_{jh}(t) = \Psi (jh + c t), \qquad \qquad \Psi (-\infty ) = 0, \qquad \Psi (+\infty ) = 1 \end{aligned}$$
(1.19)

that bifurcate from the pair \((\Psi _*, c_*)\). In particular, we study the travelling wave system

$$\begin{aligned} c \Psi '(\tau ) = {\mathcal {G}}\Big ( \{\Psi (\tau + k h)\}_{k \le 1} \Big ) \end{aligned}$$
(1.20)

posed in terms of the computational coordinate \(\tau = jh + ct \) rather than the physical coordinate \(\xi = x_j(t) + ct\) appearing in (1.5). We note that the discrete term jh now plays the role of \(\theta \).

To appreciate the advantage of this indirect approach, we note that any attempt to use \(\xi \) will lead to an equation for the waveprofile \(\Phi \) with shifts that depend on the waveprofile \(\Phi \) itself. In particular, the resulting wave equation is a state-dependent MFDE with infinite range interactions. At the moment, even state-dependent delay equations with a finite number of shifts are technically very challenging to analyze, requiring special care in the linearization procedure [41]. Indeed, linearizations typically involve higher order (continuous) derivatives, making it very hard to close fixed-point arguments.

Physical frame It turns out that there is a close relation between the two wave Ansatzes (1.5) and (1.19). In order to see this, let us assume for the moment that we have found a triplet \((\Phi , c , x)\) for which x and the function U defined in (1.5) satisfy (1.2) together with (1.3)–(1.4). Let us also assume that for each \(\vartheta \in {\mathbb {R}}\) there is a unique increasing sequence \(y_{jh;\vartheta }\) with \(y_{0;\vartheta } = \vartheta \) for which

$$\begin{aligned} \big (\Phi (y_{(j+1)h;\vartheta } ) - \Phi (y_{jh;\vartheta }) \big )^2 + (y_{(j+1)h;\vartheta } - y_{jh;\vartheta })^2 = h^2 \end{aligned}$$
(1.21)

holds for all \(j \in {\mathbb {Z}}\). This can be arranged by imposing a-priori Lipschitz bounds on \(\Phi \) and \(\Phi '\) and picking \(h > 0\) to be sufficiently small. Finally, let us assume for definiteness that \(c > 0\) and that the wave outruns the grid in the sense that \({\dot{x}}_0(t) + c> \epsilon > 0\).

A direct consequence of this inequality is that

$$\begin{aligned} x_{0}(T) + c T = x_h(0) \end{aligned}$$
(1.22)

for some \(T > 0\), which implies

$$\begin{aligned} U_0(T) = U_h(0) = \Phi \big ( x_h(0) \big ). \end{aligned}$$
(1.23)

The uniqueness property discussed above hence implies that

$$\begin{aligned} U_{jh}(T) =\Phi ( y_{jh; x_h(0) } ) = \Phi \big ( x_{(j+1)h} (0 ) \big ) \end{aligned}$$
(1.24)

for all \(j \in {\mathbb {Z}}\). Since

$$\begin{aligned} \begin{array}{lcl} \big (x_{(j+1)h}(T) - x_{jh}(T) \big )^2 &{} = &{} h^2 - \big (U_{(j+1)h}(T) - U_{jh}(T)\big )^2 \\ &{} = &{} h^2 - \Big (\Phi \big ( x_{(j+2)h}(0) \big ) - \Phi \big ( x_{(j+1)h }(0) \big ) \Big )^2 \\ &{} = &{} \big (x_{(j+2)h}(0) -x_{(j+1)h}(0) \big )^2, \\ \end{array} \end{aligned}$$
(1.25)

we see that in fact

$$\begin{aligned} x_{jh}(T) + c T = x_{(j+1)h}(0) \end{aligned}$$
(1.26)

for all \(j \in {\mathbb {Z}}\). Taking the limit \(j \rightarrow - \infty \), the boundary conditions (1.4) imply that \(c T = h\). Exploiting the well-posedness of our dynamics in forward and backward time, we conclude that

$$\begin{aligned} x_{jh}(t) = x_{0}(jT + t) + j h \end{aligned}$$
(1.27)

holds for all \(j \in {\mathbb {Z}}\) and \(t \in {\mathbb {R}}\). Writing \(\Xi (\vartheta ) = x_0(\vartheta / c)\), we hence find

$$\begin{aligned} x_{jh}(t) - j h = \Xi ( j h + c t ), \end{aligned}$$
(1.28)

which implies that

$$\begin{aligned} U_{jh}(t) = \Phi (x_{jh}(t) + ct ) = \Phi \big ( jh + \Xi (j h + ct ) + ct \big ) \end{aligned}$$
(1.29)

for all \(j \in {\mathbb {Z}}\) and \(t \in {\mathbb {R}}\). Upon introducing the function

$$\begin{aligned} \Psi (\tau ) = \Phi \big (\tau + \Xi (\tau ) \big ), \end{aligned}$$
(1.30)

this allows us to obtain the representation

$$\begin{aligned} \big ( U_{jh}(t), x_{jh}(t) - jh \big ) = \big ( \Psi (jh + ct) , \Xi (jh + ct) \big ). \end{aligned}$$
(1.31)

In fact, we show that for arbitrary solutions U to (1.15) for which \(U(t_0)\) is close to \(\Psi _*(h{\mathbb {Z}} + \vartheta )\), we indeed have the pointwise inequalities \(\left| {\dot{x}}(t_0)\right| < \left| c \right| \) whenever c is sufficiently close to \(c_*\). This can be used to show that the coordinate transformation (1.30) can be inverted, allowing us to reconstruct the profile \(\Phi (\xi )\) from \(\Psi (\tau )\).

Fixed-point setup In order to construct our travelling waves, we write \(\Psi = \Psi _* + v\) and decompose (1.20) into the form

$$\begin{aligned} \begin{array}{lcl} (c -c_*) v'= & {} {\mathcal {L}}_h v + {\mathcal {G}}_{\mathrm {nl}} \Big ( \{ v(\tau + k h ) \}_{k \le 1} \Big ) + {\mathcal {G}}\big ( \{ \Psi _*( \tau + k h) \}_{k \le 1} \big ) - c \Psi _*', \end{array} \end{aligned}$$
(1.32)

using the linear operators \({\mathcal {L}}_h\) that were introduced in [29]. In the limit \(h \downarrow 0\), these operators reduce formally to the operator \({\mathcal {L}}_*\) associated to the linearization (1.17) around the wave (1.18). The singular nature of the transition between (1.15) and (1.17) is fully encoded in the transition between \({\mathcal {L}}_*\) and \({\mathcal {L}}_h\), which was studied at length in [29]. As a result, our analysis in this paper can be seen as the construction of a regular fixed point problem. However, there are two main challenges that need to be overcome.

Fig. 1
figure 1

Plots of the initial condition and approximate solution at \(t=40\) to the discretized PDE (1.2) using the cubic nonlinearity (1.6) with \(a = 0.8536\). This bistable reaction–diffusion equation was coupled to the moving mesh system (1.33) with \(\sigma = 10^{-3}\) and \(\alpha = 10^3\), which aims to equidistribute a modified arclength monitor function. The red markers represent the location of the gridpoints, which become denser in areas where the solution profile is steep. The problem was solved on a finite domain (0, 50) with \(N=51\) grid points and (inhomogeneous) Dirichlet boundary conditions

The first complication is that \({\mathcal {L}}_h\) is not the ‘exact’ linearization of \({\mathcal {G}}\), which is far too complicated to handle. Instead, we recover our operator \({\mathcal {L}}_h\) after several simplification steps, which each introduce h-dependent errors that need to kept under control. In order to achieve this, we reapply the approximation framework developed in [28] in order to systematically bound the global errors that arise by modifying the individual factors of the products that appear in the definition of \({\mathcal {G}}\).

The second obstacle is that the nonlinearity \({\mathcal {G}}\) acts on sequences \(U: h{\mathbb {Z}}\rightarrow {\mathbb {R}}\), while the fixed-point problem (1.32) is formulated in terms of functions \(v \in L^2\). Since the relevant transitions between supremum and \(L^2\)-based norms cost a factor of \(h^{-1/2}\), special care must be taken to construct appropriate function spaces that allow uniform bounds for \(h \downarrow 0\). This is particularly dangerous for the terms that are quadratic in the second differences of U, which correspond roughly to the \(\int _- \gamma ^{-4} w_{\theta \theta }^2\) term in (1.17).

In fact, we need to exploit the special structure of \({\mathcal {G}}\) and take a discrete derivative of (1.32) in order to close our problem. We hence need to obtain estimates on the discrete derivative of the nonlinear residual \({\mathcal {G}}_{\mathrm {nl}}\), which requires an elaborate bookkeeping system.

Outlook We view our work here as a first step towards understanding the impact of adaptive discretization schemes on travelling waves and other patterns that exist for all time. In particular, we believe that the waves constructed here can be seen as a slow manifold for the dynamics of the full system (1.2) with the non-instantaneous gridpoint behaviour

$$\begin{aligned} \begin{array}{lcl} \sigma {\dot{x}}_{jh} &{} = &{} \sqrt{ (x_{(j+1)h} - x_{jh})^2 + \alpha (U_{(j+1)h} - U_{jh})^2 }\\ &{}&{}\quad - \sqrt{ (x_{(j-1)h} - x_{jh})^2 + \alpha (U_{(j-1)h} - U_{jh})^2 } \end{array} \end{aligned}$$
(1.33)

prescribed by the MMPDE5 scheme [22] with \(\alpha =1\). Here \(\sigma > 0\) is a tunable speed parameter, which we effectively set to zero by passing to (1.3).

Fig. 2
figure 2

Plots of the temporal evolution of the mesh locations and solution values for the system discussed in Fig. 1. Notice that both quantities display the wave-like behaviour (1.31) after an initial transient has passed

Using the Fredholm theory developed in [29] for the operators \({\mathcal {L}}_h\) one should be able to leverage the ideas in [40] to effectively track the fast grid-dynamics in the \(0 < \sigma \ll 1\) regime. A further step in the program would be to also handle temporal discretizations, inspired by the approach developed in [27] that we described above. Finally, we feel that it is important to understand the stability of the discretized waves under the full dynamics of the numerical scheme. To achieve this, one could follow the approach in [39] to transfer information from the operators \({\mathcal {L}}_h\) to the linearization around the actual adaptive travelling waves constructed in this paper.

To illustrate the feasability of this program, we used the MMPDELab package [20] to numerically compute the solution of the discretized system (1.2) coupled to the moving mesh scheme (1.33); see Figs. 1 and 2. Upon initializing the problem with the exact wave solution to the PDE (1.1), the system quickly converged to a modified state whereby both the location of the gridpoints and the associated solution values displayed wave-like behaviour.

We are specially interested here in the pinning phenomenon. Indeed, numerical observations indicate that the set of detuning parameters a for which \(c(a) = 0\) shrinks dramatically when using adaptive discretizations. Understanding this in a rigorous fashion would give considerable insight into the theoretical benefits of adaptive grids compared to the practical benefits of increased performance. Preliminary results in this direction can be found in [21].

Let us emphasize that the application range of our techniques does not appear to be restricted to the scalar problem (1.1) or the specific grid-update scheme (1.33) that we use. Indeed, using the framework developed in [39], it should be possible to perform a similar analysis for the FitzHugh-Nagumo equation PDE and other multi-component reaction–diffusion problems. In addition, any numerical scheme based on the arclength monitor function will share (1.3) as the instantaneous equidistribution limit.

Overview This paper is organized as follows. After formulating our main results in Sect. 2, we introduce our notational framework and recap the key contributions from [28, 29] in Sect. 3. In Sect. 4 we simplify the nonlinear functions that appear as factors in the product structure of \({\mathcal {G}}\) and obtain estimates on all the resulting errors. These estimates are used in Sect. 5 to compute tractable expressions for the linearization of \({\mathcal {G}}\) and its discrete derivative \({\mathcal {G}}^+\) around \(\Psi _*\) and obtain bounds on the residuals. We conclude in Sect. 6 by combining all these ingredients with the theory developed in [29]. In particular, we develop an appropriate fixed-point argument to construct our desired travelling waves.

In order to develop the main story in a reasonably streamlined fashion that focuses on the key ideas, we have chosen to transfer many of the tedious underlying estimates and algebraic manipulations to the appendices. In order to keep this paper as self-contained as possible, these appendices also summarize some of the fundamental auxiliary bounds that were obtained in [28, 29].

2 Main Results

The main results of this paper concern adaptive-grid discretizations of the scalar PDE

$$\begin{aligned} u_t = u_{xx} + g(u). \end{aligned}$$
(2.1)

Throughout the paper, we assume that the nonlinearity g satisfies the following standard bistability condition.

  1. (Hg)

    The nonlinearity \(g: {\mathbb {R}}\rightarrow {\mathbb {R}}\) is \(C^3\)-smooth and has a bistable structure, in the sense that there exists a constant \(0< a < 1\) such that we have

    $$\begin{aligned} g(0) = g(a) = g(1) = 0, \qquad g'(0)< 0, \qquad g'(1) < 0, \end{aligned}$$
    (2.2)

    together with

    $$\begin{aligned} g(u) < 0 \hbox { for } u \in (0, a) \cup (1, \infty ), \qquad g(u) > 0 \hbox { for } u \in (-\infty ,-1) \cup (a, 1). \end{aligned}$$
    (2.3)

It is well-known that the PDE (2.1) admits a travelling wave solution that connects the two stable equilibria of g [15]. The key requirement in our next assumption is that this wave is not stationary, which can be arranged by demanding \(\int _0^1 g(u) \, du \ne 0\).

(H\(\Phi _*\)):

There exists a wave speed \(c_* \ne 0\) and a profile \(\Phi _* \in C^5({\mathbb {R}}, {\mathbb {R}})\) that satisfies the limits

$$\begin{aligned} \lim _{\xi \rightarrow - \infty } \Phi _*(\xi ) = 0, \qquad \lim _{\xi \rightarrow + \infty } \Phi _*(\xi ) = 1 \end{aligned}$$
(2.4)

and yields a solution to the PDE (2.1) upon writing

$$\begin{aligned} u(x,t) = \Phi _*( x + c_* t). \end{aligned}$$
(2.5)

In [28] we derived an effective equation for the dynamics of the sequence \(U(t): h {\mathbb {Z}}\rightarrow {\mathbb {R}}\) featuring in the adaptive scheme (1.2)–(1.4) for (2.1) that no longer explictly depends on the location of the gridpoints. In order to formulate this reduced equation, we recall the discrete derivatives

$$\begin{aligned} {[}\partial ^+ U]_{jh}= & {} h^{-1}\big [ U_{(j+1)h} - U_{jh} \big ], \nonumber \\ {[}\partial ^- U]_{jh}= & {} h^{-1}\big [ U_{jh} - U_{(j-1)h} \big ], \nonumber \\ {[}\partial ^0 U]_{jh}= & {} (2h)^{-1}\big [ U_{(j+1)h} - U_{(j-1)h} \big ], \end{aligned}$$
(2.6)

together with the first-order differences

$$\begin{aligned} {\mathcal {D}}^{\diamond _\pm }(U) = \frac{\partial ^\pm U}{\sqrt{1 - (\partial ^\pm U)^2 }}, \qquad \qquad {\mathcal {D}}^{\diamond _0}(U) = \frac{2 \partial ^0 U}{ \sqrt{1 - (\partial ^+ U)^2 } + \sqrt{1 - (\partial ^- U)^2 } } \end{aligned}$$
(2.7)

and the second order analogues

$$\begin{aligned} {\mathcal {D}}^{\diamond \diamond _0} (U) = \frac{2}{h}\frac{ {\mathcal {D}}^{\diamond _+}(U) - {\mathcal {D}}^{\diamond _-}(U) }{ \sqrt{1 - (\partial ^+ U)^2 } + \sqrt{1 - (\partial ^- U)^2 } }, \qquad \qquad {\mathcal {D}}^{\diamond _0; +}(U) = \partial ^+ {\mathcal {D}}^{\diamond _0}(U) . \end{aligned}$$
(2.8)

This allows us to introduce the auxiliary functions

$$\begin{aligned} p(U) = \frac{{\mathcal {D}}^{\diamond _+}(U) }{1 + {\mathcal {D}}^{\diamond _+}(U) {\mathcal {D}}^{\diamond _0}(U) }, \qquad \qquad q(U) = h^{-1} \ln \big [ 1 + h p(U) {\mathcal {D}}^{\diamond _0;+}(U) \big ], \end{aligned}$$
(2.9)

which using the notation

$$\begin{aligned} \left[ \sum _{-;h} a \right] _{jh} = h \sum _{k> 0} a_{(j - k) h }, \qquad \qquad \left[ \sum _{+;h} a \right] _{jh} = h \sum _{k > 0} a_{(j + k) h } \end{aligned}$$
(2.10)

allows us to recall the definitions

$$\begin{aligned} {\mathcal {Q}}(U) = \sum _{-;h} q(U) , \qquad \qquad {\mathcal {Z}}^\pm (U) = \mathrm {exp}\big [ \pm {\mathcal {Q}}(U) \big ] \end{aligned}$$
(2.11)

and subsequently write

$$\begin{aligned} {\mathcal {G}}(U)= & {} {\mathcal {D}}^{\diamond \diamond _0}\big ( U \big ) + g \big ( U \big )\nonumber \\&-{\mathcal {D}}^{\diamond _0}\big (U\big ) {\mathcal {Z}}^-(U) \sum _{-;h} p(U) {\mathcal {Z}}^+(U) \partial ^+\big [ {\mathcal {D}}^{\diamond \diamond _0}(U) + g(U) \big ] . \end{aligned}$$
(2.12)

These ingredients allow us to formulate the effective reduced system [28,  Eq. (2.25)] for the dynamics of (1.2)–(1.4) as

$$\begin{aligned} {\dot{U}}(t) = {\mathcal {G}}\big (U(t) \big ) , \end{aligned}$$
(2.13)

which will be the main system that we analyze in this paper.

We recall the arclength parametrization \(\xi _*(\tau )\) defined by the identity

$$\begin{aligned} {\mathcal {A}}\big (\xi _*(\tau )\big ) = \int _0^{\xi _*(\tau )} \sqrt{1 + [\partial _{\xi '} \Phi _*(\xi ')]^2 } \, d \xi ' = \tau , \end{aligned}$$
(2.14)

together with the stretched waveprofile \(\Psi _*: {\mathbb {R}}\rightarrow {\mathbb {R}}\) given by

$$\begin{aligned} \begin{array}{lcl} \Psi _*(\tau ) = \Phi _*\big ( \xi _*(\tau ) \big ). \end{array} \end{aligned}$$
(2.15)

The main result of this paper states that for sufficiently small \(h > 0\), the reduced problem (2.13) admits a travelling wave solution

$$\begin{aligned} U_{jh}(t) = \Psi _h( jh + c_h t) \end{aligned}$$
(2.16)

with \((\Psi _h, c_h) \approx (\Psi _*, c_*)\) in an appropriate sense. These waves are locally unique up to translation. We note that items (iv) and (v) use the notation \(\partial ^+_h v = h^{-1} [v(\cdot + h) -v(\cdot ) ]\) for functions v. In addition, we use the shorthands \(L^2 = L^2({\mathbb {R}};{\mathbb {R}})\) and \(H^1 = H^1({\mathbb {R}}; {\mathbb {R}})\), together with the Heaviside sequence \(H_{jh} = {\mathbf {1}}_{j \ge 0}\).

Theorem 2.1

(see Sect. 6) Suppose that (Hg) and (H\(\Phi _*\)) are satisfied. Then there exists a constant \(\delta _h > 0\) together with pairs

$$\begin{aligned} (\Psi _h , c_h) \in C^1({\mathbb {R}}; {\mathbb {R}}) \times {\mathbb {R}}, \end{aligned}$$
(2.17)

defined for \(0 < h \le \delta _h\), such that the following properties are satisfied.

  1. (i)

    For every \(0 < h \le \delta _h\) we have the limits

    $$\begin{aligned} \lim _{\xi \rightarrow - \infty } \Psi _h(\xi ) = 0, \qquad \qquad \lim _{\xi \rightarrow + \infty } \Psi _h(\xi ) = 1. \end{aligned}$$
    (2.18)
  2. (ii)

    For every \(0 < h \le \delta _h\) we have the strict inequality

    $$\begin{aligned} \sup _{\tau \in {\mathbb {R}}} \left| \Psi _h( \tau + h) - \Psi _h(\tau ) \right| < h. \end{aligned}$$
    (2.19)
  3. (iii)

    For every \(0 < h \le \delta _h\), the function \(U: {\mathbb {R}}\rightarrow \ell ^\infty (h{\mathbb {Z}}; {\mathbb {R}})\) defined by

    $$\begin{aligned} U_{jh}(t) = \Psi _h(jh + c_h t) \end{aligned}$$
    (2.20)

    satisfies the inclusion

    $$\begin{aligned} t \mapsto U(t) - H \in C^1\big ({\mathbb {R}}; \ell ^2(h{\mathbb {Z}}; {\mathbb {R}}) \big ). \end{aligned}$$
    (2.21)

    In addition, the identity (2.13) and the strict inequality \(\left\| \partial ^+ U(t) \right\| _\infty < 1\) both hold for all \(t \in {\mathbb {R}}\).

  4. (iv)

    We have \(\Psi _h - \Psi _* \in H^1\) for every \(0 < h \le \delta _h\) and the limit

    $$\begin{aligned} \begin{array}{lcl} \left| c_h - c_* \right| + \left\| \Psi _h - \Psi _* \right\| _{H^1} + \left\| \partial ^+_h \big [ \Psi _h - \Psi _* \big ] \right\| _{H^1} \\ \quad + \left\| \partial ^+_h \partial ^+_h \partial ^+_h \big [ \Psi _h - \Psi _* ]\right\| _{L^2} \rightarrow 0 \end{array} \end{aligned}$$
    (2.22)

    holds as \(h \downarrow 0\).

  5. (v)

    Pick any \(0 < h \le \delta _h\) and consider a pair \(({\widetilde{\Psi }} , {\tilde{c}}) \in L^\infty \times {\mathbb {R}}\) that has \({\widetilde{\Psi }} - \Psi _* \in H^1\) with

    $$\begin{aligned} \left| {\tilde{c}} - c_*\right| + \left\| {\widetilde{\Psi }} - \Psi _* \right\| _{H^1} + \left\| \partial ^+_h \big [{\widetilde{\Psi }} - \Psi _* \big ] \right\| _{H^1}\nonumber \\ + \left\| \partial ^+_h \partial ^+_h \partial ^+_h \big [{\widetilde{\Psi }} - \Psi _* \big ] \right\| _{L^2} < h^{3/4}. \end{aligned}$$
    (2.23)

    Then the function \({\widetilde{U}}: {\mathbb {R}}\rightarrow \ell ^\infty (h{\mathbb {Z}}; {\mathbb {R}})\) defined by

    $$\begin{aligned} {\widetilde{U}}_{jh}(t) = {\widetilde{\Psi }}_h\big (jh + {\tilde{c}} t\big ) \end{aligned}$$
    (2.24)

    satisfies the inclusion

    $$\begin{aligned} t \mapsto {\widetilde{U}}(t) - H \in C^0\big ({\mathbb {R}}; \ell ^2(h{\mathbb {Z}}; {\mathbb {R}}) \big ), \end{aligned}$$
    (2.25)

    together with the strict inequality \(\left\| \partial ^+ {\widetilde{U}}\right\| _\infty < 1\) for all \( t \in {\mathbb {R}}\). In addition, if \({\widetilde{U}}\) is a solution to the system (2.13) for all \(t \in {\mathbb {R}}\), then we must have

    $$\begin{aligned} \big ({\widetilde{\Psi }}(\cdot ),{\tilde{c}} \big ) = \big (\Psi _h(\cdot + \vartheta ), c_h \big ) \end{aligned}$$
    (2.26)

    for some \(\vartheta \in {\mathbb {R}}\).

We emphasize that the location of the gridpoints for the waves (2.16) can be determined by using

$$\begin{aligned} x_{jh}(t)= & {} jh - \sum _{j' < j} \frac{ \big (U_{(j'+1)h}(t) - U_{j'h}(t) \big )^2 }{ \sqrt{ h^2 - (U_{(j'+1)h}(t) - U_{j'h}(t))^2 } + h }; \end{aligned}$$
(2.27)

see [28,  Thm. 2.3]. In fact, our final result shows how these waves in the computational coordinates can be interpreted as wave-like objects in the original physical coordinates.

Corollary 2.2

(see Sect. 6) Consider the setting of Theorem 2.1. Then there exists a constant \(0<{\widetilde{\delta }}_h < \delta _h\) so that for all \(0 < h \le {\widetilde{\delta }}_h\) there exist pairs

$$\begin{aligned} (\Xi _h , \Phi _h) \in C^1({\mathbb {R}}; {\mathbb {R}}) \times C^1({\mathbb {R}}; {\mathbb {R}}) \end{aligned}$$
(2.28)

that satisfy the following properties.

  1. (i)

    Upon writing

    $$\begin{aligned} \begin{array}{lcl} x_{jh}(t) &{} = &{} jh + \Xi _h(jh + c_h t ), \\ U_{jh}(t) &{} = &{} \Psi _h(jh + c_h t) , \end{array} \end{aligned}$$
    (2.29)

    the adaptive grid equations (1.2)–(1.4) are satisfied for all \(t \in {\mathbb {R}}\).

  2. (ii)

    For every \(t \in {\mathbb {R}}\) and \(j \in {\mathbb {Z}}\), the functions defined in (2.29) satisfy the relation

    $$\begin{aligned} U_{jh}(t) = \Phi _h\big (x_{jh}(t) + c_h t \big ). \end{aligned}$$
    (2.30)

We remark that if (2.16) and (2.30) both hold, simple substitutions yield the identity

$$\begin{aligned} \Psi _h(jh + c_h t)= & {} U_{jh}(t) \nonumber \\= & {} \Phi _h\big ( jh + \Xi _h(jh + c_h t) + c_h t \big ) \nonumber \\= & {} \Phi _h( jh + c_h t + \Xi _h(jh + c_h t) \big ). \end{aligned}$$
(2.31)

In particular, the main assertion in Corollary 2.2 is that the perturbed coordinate transformation

$$\begin{aligned} \xi _h(\tau ) =\tau + \Xi _h(\tau ) \end{aligned}$$
(2.32)

is invertible for sufficiently small \(h > 0\), allowing us to transfer the waves back to the original physical framework.

3 Setup and Notation

In this section we recall several crucial results and notational conventions introduced in the prequel papers [28, 29]. This will ensure that the current paper can be read reasonably independently. As a preparation, we recall the sequence spaces

$$\begin{aligned} \ell ^2_h= & {} \{ V: h {\mathbb {Z}}\rightarrow {\mathbb {R}}\hbox { for which } \left\| V\right\| _{\ell ^2_h}^2 := h \sum _{j \in {\mathbb {Z}}} \left| V_{hj} \right| ^2< \infty \} , \nonumber \\ \ell ^\infty _h= & {} \{ V: h {\mathbb {Z}}\rightarrow {\mathbb {R}}\hbox { for which } \left\| V\right\| _{\ell ^\infty _h} := \sup _{j \in {\mathbb {Z}}} \left| V_{hj} \right| < \infty \} \end{aligned}$$
(3.1)

that were introduced in [28,  §3.3], together with the higher order norms

$$\begin{aligned} \left\| V\right\| _{\ell ^{2;1}_h}= & {} \left\| V\right\| _{\ell ^2_h} + \left\| \partial ^+V\right\| _{\ell ^2_h} ,\nonumber \\ \left\| V\right\| _{\ell ^{2;2}_h}= & {} \left\| V\right\| _{\ell ^2_h} + \left\| \partial ^+V\right\| _{\ell ^2_h} + \left\| \partial ^+ \partial ^+ V\right\| _{\ell ^2_h} , \nonumber \\ \left\| V\right\| _{\ell ^{2;3}_h}= & {} \left\| V\right\| _{\ell ^2_h} + \left\| \partial ^+V\right\| _{\ell ^2_h} + \left\| \partial ^+ \partial ^+ V\right\| _{\ell ^2_h} + \left\| \partial ^+ \partial ^+ \partial ^+ V\right\| _{\ell ^2_h } \end{aligned}$$
(3.2)

and their counterparts

$$\begin{aligned} \left\| V\right\| _{\ell ^{\infty ;1}_h}= & {} \left\| V\right\| _{\ell ^\infty _h} + \left\| \partial ^+V\right\| _{\ell ^\infty _h} , \nonumber \\ \left\| V\right\| _{\ell ^{\infty ;2}_h}= & {} \left\| V\right\| _{\ell ^\infty _h} + \left\| \partial ^+V\right\| _{\ell ^\infty _h} + \left\| \partial ^+ \partial ^+ V\right\| _{\ell ^\infty _h} . \end{aligned}$$
(3.3)

For a single fixed \(h > 0\) all these norms are naturally equivalent to the \(\ell ^2_h\)-norm or the \(\ell ^\infty _h\)-norm. The point here is that the \(h^{-1}\) factor in the definition of \(\partial ^+\) introduces a natural scaling that will allow us to formulate h-independent bounds.

In addition, we pick a reference function \(U_{\mathrm {ref};*} \in C^2({\mathbb {R}}, [0,1])\) that satisfies the properties

$$\begin{aligned} U_{\mathrm {ref};*}\big ((-\infty , -2]\big ) = 0, \qquad U_{\mathrm {ref};*}\big ([2, \infty )\big ) = 1, \qquad 0 \le U'_{\mathrm {ref};*}< 1 , \qquad \left| U''_{\mathrm {ref};*} \right| < 1.\nonumber \\ \end{aligned}$$
(3.4)

For any \(\kappa > 0\), this allows us to write

$$\begin{aligned} U_{\mathrm {ref};\kappa }(\tau ) = U_{\mathrm {ref};*}(\kappa \tau ) \end{aligned}$$
(3.5)

and introduce the subset

$$\begin{aligned} {\mathcal {V}}_{h; \kappa }= & {} \{ V \in \ell ^2_h: \left\| V\right\| _{\ell ^{2;2}_h} + \left\| V\right\| _{\ell ^\infty _h} + \left\| \partial ^+ \partial ^+ V\right\| _{\ell ^\infty _h}< \frac{1}{2}\kappa ^{-1} \hbox { and } \nonumber \\&\left\| \partial ^+ V \right\| _{\ell ^\infty _h} < 1 - 2\kappa \} , \end{aligned}$$
(3.6)

which is open in \(\ell ^2_h\) on account of the continuity of the difference operator and the supremum norm. We can now recall the affine subset [28,  §3.4]

$$\begin{aligned} \Omega _{h;\kappa } = U_{\mathrm {ref};\kappa }(h{\mathbb {Z}}) + {\mathcal {V}}_{h;\kappa } \subset \ell ^\infty _h \end{aligned}$$
(3.7)

that captures the admissable states of the waves that we are interested in and provides adequate control on the necessary difference operators.

Indeed, for each \(U \in \Omega _{h;\kappa }\) we have the crucial bound \(\left\| \partial ^+U\right\| _{\ell ^\infty _h} \le 1 - \kappa \), which ensures that our grid points are well-defined. In addition, the norms \(\left\| \partial ^+ U\right\| _{\ell ^{2,1}_h}\), \(\left\| U\right\| _{\ell ^{\infty ;2}_h}\) and \(\left\| g(U)\right\| _{\ell ^2_h}\) are all bounded uniformly in \(h > 0\). Finally, it is possible to pick \(\epsilon _0 > 0\) and \(\kappa > 0\) in such a way that for any \(0< h < 1\) and any \(v \in H^1\) that has

$$\begin{aligned} \left\| v\right\| _{H^1} + h^{-1/2} \left\| \partial ^+ v \right\| _{H^1} < 2 \epsilon _0, \end{aligned}$$
(3.8)

we have the inclusion

$$\begin{aligned} \big [\Psi _* + v\big ](\vartheta + h {\mathbb {Z}}) \in \Omega _{h;\kappa } \end{aligned}$$
(3.9)

for all \(\vartheta \in [0,h]\). These statements all follow from [28,  Prop. 3.1-3.3]. Note that none of these estimates involve third-differences, which will always be tracked separately throughout the paper. Indeed, in Sect. 6 we explain how their special structure plays a crucial role in the proof of our main result.

3.1 Linear Operators

In [29] we analyzed several important linear operators that will turn out to be closely related to our travelling wave system (1.20). To set the stage, we recall the sequences

$$\begin{aligned} \gamma _U = \sqrt{1 - (\partial ^0 U )^2 }, \qquad \qquad \partial ^{(2)} U = \partial ^+ \partial ^- U, \end{aligned}$$
(3.10)

which are well defined for any \(U \in \Omega _{h;\kappa }\). Following [29,  §5], we introduce the linear operators \(M_U: \ell ^2_h \rightarrow \ell ^2_h\) that act as

$$\begin{aligned} M_U[V] = -c_* \gamma _U^{-1} \partial ^0 V + 4 \gamma _U^{-4} \partial ^0 U [ \partial ^{(2)} U] \partial ^0 V + \gamma _U^{-2} \partial ^{(2)} V + \gamma _U^2 g'(U) V, \end{aligned}$$
(3.11)

together with their twisted counterparts \(L_U: \ell ^2_h \rightarrow \ell ^2_h\) defined by

$$\begin{aligned} \begin{array}{lcl} L_U[V]= & {} c_* \partial ^0 V + M_U[V] + \partial ^0 U \sum _{-;h} \gamma _U^{-2} [\partial ^{(2)} U ] M_U[V] , \end{array} \end{aligned}$$
(3.12)

always taking \(U \in \Omega _{h;\kappa }\).

A special role is reserved for the discrete derivative \(\partial ^+M_U\), which we approximate by the linear operator

$$\begin{aligned} M^+_{U;\mathrm {apx}}[V] = \gamma _U^2 \big ( {\widetilde{M}}_{U;I}[V] + {\widetilde{M}}_{U;II}[V] + {\widetilde{M}}_{U;III}[V] \big ) \nonumber \\ - 2 \gamma _U^{-2} \partial ^0 U [ \partial ^{(2)} U] M_U[V]. \end{aligned}$$
(3.13)

Based on the computations in [29,  Prop. 5.5], this decomposition uses the expressions

$$\begin{aligned} {\widetilde{M}}_{U;I}[V]= & {} 4 [6 \gamma _U^{-8} - 5 \gamma _U^{-6} ] [\partial ^{(2)} U ]^2 \partial ^0 V \nonumber \\&+ 8 \gamma _U^{-6} \partial ^0 U [ \partial ^{(2)} U] \partial ^{(2)} V \nonumber \\&\qquad + g''(U) [\partial ^0 U ] V + g'(U) \partial ^0 V , \nonumber \\ {\widetilde{M}}_{U;II}[V]= & {} -3 c_* \gamma _U^{-5} \partial ^0 U [\partial ^{(2)} U] \partial ^0 V - c_* \gamma _U^{-3} \partial ^{(2)} V \end{aligned}$$
(3.14)

that feature at most second differences of V, together with

$$\begin{aligned} {\widetilde{M}}_{U;III}[V] = 4 \gamma _U^{-6} \partial ^0 U [\partial ^+ \partial ^{(2)} U ] \partial ^0 V + \gamma _U^{-4} \partial ^+ \partial ^{(2)} V \end{aligned}$$
(3.15)

which contains third differences that needs to be treated carefully. Several crucial bounds for these operators are collected in Proposition C.1.

We are now ready to recall the linear operator \({\mathcal {L}}_h: H^1 \rightarrow L^2\) that acts as

$$\begin{aligned} {\mathcal {L}}_h v= & {} - c_* v' + L_{\Psi _*} v, \end{aligned}$$
(3.16)

where we are slightly abusing notation. Indeed, recalling the discrete evaluation operator

$$\begin{aligned}{}[\mathrm {ev}_{\vartheta } f]_{jh} = f(\vartheta + jh ) \end{aligned}$$
(3.17)

that ‘samples’ a function f on the grid \(\vartheta + h {\mathbb {Z}}\), the identity \({\mathcal {L}}_h v = f\) should be interpreted as the statement that

$$\begin{aligned} \mathrm {ev}_{\vartheta } \big [ c_* v' + f \big ] = L_{\mathrm {ev}_{\vartheta } \Psi _*}[\mathrm {ev}_{\vartheta }v ] \end{aligned}$$
(3.18)

holds for each \(\vartheta \in [0,h]\). We remark that the right-hand side above is continuous in \(\ell ^2_h\) as a function of \(\vartheta \) as a consequence of (A.6) and the continuity of the translation operator on \(H^1\).

The key purpose of [29] was to construct a quasi-inverse for the operator \({\mathcal {L}}_h\). Indeed, [29,  Thm. 2.3] establishes the existence of two linear maps

$$\begin{aligned} \beta ^*_h: L^2 \rightarrow {\mathbb {R}}, \qquad \qquad {\mathcal {V}}^*_h: L^2 \rightarrow H^1, \end{aligned}$$
(3.19)

defined for small \(h >0\), so that for each \(f \in L^2\) the pair

$$\begin{aligned} (\beta , v) = \big ( \beta ^*_{h} f, {\mathcal {V}}^*_{h} f \big ) \in {\mathbb {R}}\times H^1 \end{aligned}$$
(3.20)

is the unique solution to the problem

$$\begin{aligned} {\mathcal {L}}_{h} v = f + \beta \Psi _*' \end{aligned}$$
(3.21)

up to a normalization condition that can be used to fix the phase of our constructed wave. The crucial point is that we obtain h-uniform bounds

$$\begin{aligned} \left| \beta ^*_{h} f \right| + \left\| {\mathcal {V}}^*_{h} f \right\| _{H^1} + \left\| \partial ^+_h \partial ^+_h {\mathcal {V}}^*_{h} f \right\| _{L^2}\le & {} K \left\| f\right\| _{L^2}, \nonumber \\ \left\| \partial ^+_h {\mathcal {V}}^*_{h} f \right\| _{H^1} + \left\| \partial ^+_h \partial ^+_h \partial ^+_h {\mathcal {V}}^*_{h} f \right\| _{L^2}\le & {} K \big [ \left\| f\right\| _{L^2} + \left\| \partial ^+_h f\right\| _{L^2} \big ], \end{aligned}$$
(3.22)

which will provide the required control on the second and third differences of our travelling wave.

We remark that these difference operators cannot be replaced by the corresponding derivatives, which forces us to develop a rather delicate fixed point argument in Sect. 6. In addition, we note that the spectral convergence framework used to obtain (3.22) relies strongly on the inner product structure of \(L^2\), which explains why we do not have \(L^\infty \)-based estimates (yet). This is the reason that we go to great lengths throughout the paper to work with \(\ell ^2_h\)-bounds as much as possible. Indeed, the results in Sect. A show that these mix well with \(L^2\)-functions, unlike supremum bounds.

3.2 Error Functions

The errors that need to be controlled during our reduction steps arise from various sources that we briefly discuss here. As preparation, we recall the translation operators

$$\begin{aligned}{}[T^+ a ]_{jh} = a_{(j+1)h}, \qquad [T^- a]_{jh} = a_{(j-1)h} \end{aligned}$$
(3.23)

and the sums and products

$$\begin{aligned} S^\pm a = \frac{1}{2}(a + T^\pm a), \qquad \qquad P^\pm a = a T^\pm a. \end{aligned}$$
(3.24)

These allow us to recall the function

$$\begin{aligned} \begin{array}{lcl} {\mathcal {E}}_{\mathrm {sm}}(U) = h \partial ^- \Big [ \gamma _U^{-4} (2 - \gamma _U^2) S^+[\partial ^{(2)} U] \Big ] \end{array} \end{aligned}$$
(3.25)

from [28,  Eq. (7.28)], which measures the smoothness of U in some sense. Indeed, it becomes small whenever third differences of U can be controlled, which is the case when taking \(U = \Psi _*\).

In a similar vein, we introduce the error functions

$$\begin{aligned} {\mathcal {E}}_{\mathrm {sh};U}(V)= & {} h \left\| V\right\| _{\ell ^{2;2}_h} , \nonumber \\ \overline{{\mathcal {E}}}_{\mathrm {sh};U}(V)= & {} h \left\| V\right\| _{\ell ^{3;2}_h} + h \big [ \left\| \partial ^+ \partial ^+ \partial ^+ U\right\| _{\ell ^2_h} + \left\| \partial ^+ \partial ^+ \partial ^+ U\right\| _{\ell ^\infty _h} \big ] \left\| V\right\| _{\ell ^{2;2}_h} , \end{aligned}$$
(3.26)

which can be used to ‘shift’ function evaluations back and forth between neighbouring lattice sites. We note in general that overlined symbols will be used for expressions related to \({\mathcal {G}}^+\), which naturally involve higher order differences than those related to \({\mathcal {G}}\). Indeed, the nonlinearities in our problem will be controlled by the product

$$\begin{aligned} {\mathcal {E}}_{\mathrm {prod}}(W^{(1)}, W^{(2)})= & {} \left\| W^{(1)}\right\| _{\ell ^{2;2}_h} \left\| W^{(2)} \right\| _{\ell ^{2;2}_h} \nonumber \\&+ \left\| W^{(1)}\right\| _{\ell ^{2;2}_h} \left\| W^{(2)}\right\| _{\ell ^{\infty ;1}_h} + \left\| W^{(1)}\right\| _{\ell ^{\infty ;1}_h} \left\| W^{(2)}\right\| _{\ell ^{2;2}_h} , \end{aligned}$$
(3.27)

together with

$$\begin{aligned} \overline{{\mathcal {E}}}_{\mathrm {prod};U}(W^{(1)}, W^{(2)} )= & {} \left\| \partial ^+ \partial ^+\! \partial ^+ \!U\right\| _{\ell ^\infty _h} \Big [\left\| W^{(1)}\right\| _{\ell ^{2;1}_h } \left\| W^{(2)}\right\| _{\ell ^{\infty ;1}_h } \!+\! \left\| W^{(1)}\right\| _{\ell ^{\infty ;1}_h } \left\| W^{(2)}\right\| _{\ell ^{2;1}_h } \Big ] \nonumber \\&\quad + \left\| W^{(1)}\right\| _{\ell ^{2;2}_h} \left\| W^{(2)} \right\| _{\ell ^{2;2}_h} \nonumber \\&\quad + \left\| W^{(1)}\right\| _{\ell ^{2;2}_h} \left\| W^{(2)}\right\| _{\ell ^{\infty ;2}_h} + \left\| W^{(1)}\right\| _{\ell ^{\infty ;2}_h} \left\| W^{(2)}\right\| _{\ell ^{2;2}_h} \nonumber \\&\quad + \left\| W^{(1)}\right\| _{\ell ^{2;3}_h} \left\| W^{(2)}\right\| _{\ell ^{\infty ;1}_h} + \left\| W^{(1)}\right\| _{\ell ^{\infty ;1}_h} \left\| W^{(2)}\right\| _{\ell ^{2;3}_h} . \end{aligned}$$
(3.28)

Observe here that the supremum norms are always at least one order smaller than the highest \(\ell ^2\)-based norms. In addition, there are no squares of third-differences or products involving only supremum bounds. These facts will turn out to be crucial when passing between sequences and functions in order to apply the estimates (3.22) in Sect. 6.

Our final error functions are given by

$$\begin{aligned} {\mathcal {E}}_{\mathrm {tw}}(U)= & {} \gamma _U^{-4} \partial ^{(2)} U + g(U) - c_* \gamma _U^{-1} \partial ^0 U , \end{aligned}$$
(3.29)

together with its approximate first difference

$$\begin{aligned} {\mathcal {E}}^+_{\mathrm {tw};\mathrm {apx}}(U)= & {} 4 \gamma _U^{-6} \partial ^0 U S^+[\partial ^{(2)} U] T^+[\partial ^{(2)} U] + \gamma _U^{-4} \partial ^+ \partial ^{(2)} U \nonumber \\&+ g'(U)\partial ^0 U - c_* \gamma _U^{-3} S^+[\partial ^{(2)} U]. \end{aligned}$$
(3.30)

The relation between these two functions is explored in Proposition C.2. We view both expressions as a measure for the difference between U and the stretched travelling wave \(\Psi _*\). Indeed, upon introducing the notation

$$\begin{aligned} \gamma _*(\tau ) = \sqrt{1 - \Psi _*' (\tau )^2 }, \end{aligned}$$
(3.31)

we recall from [29,  Eq. (3.4)] that \(\Psi _*\) satisfies the ODE

$$\begin{aligned} c_* \gamma _*^{-1} \Psi _*'= & {} \gamma _*^{-4} \Psi _*'' + g\big (\Psi _*\big ), \end{aligned}$$
(3.32)

which resembles the continuum limit of (3.29). This can be differentiated to yield

$$\begin{aligned} c_* \gamma _*^{-3}\Psi _*''= & {} \gamma _*^{-4} \Psi _*''' + 4 \gamma _*^{-6} \Psi _*' (\Psi _*'')^2 + g'\big (\Psi _*\big ) \Psi _*', \end{aligned}$$
(3.33)

the natural limit of (3.30).

Together with the smoothness term \({\mathcal {E}}_{\mathrm {sm}}\), the functions (3.29) and (3.30) can be used to define our final remainder terms

$$\begin{aligned} {\mathcal {E}}_{\mathrm {rem};U}(V)= & {} \left\| V\right\| _{\ell ^{2;2}_h} \Big [ \left\| {\mathcal {E}}_{\mathrm {tw}}(U) \right\| _{\ell ^2_h} + \left\| {\mathcal {E}}_{\mathrm {tw}}(U) \right\| _{\ell ^\infty _h} + \left\| {\mathcal {E}}_{\mathrm {sm}}(U) \right\| _{\ell ^2_h} \Big ] , \nonumber \\ \overline{{\mathcal {E}}}_{\mathrm {rem};U}(V)= & {} \left\| V\right\| _{\ell ^{2;2}_h} \Big [ \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^2_h} + \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^\infty _h} + \left\| {\mathcal {E}}_{\mathrm {sm}}(U)\right\| _{\ell ^2_h} \Big ] \nonumber \\&+ \left\| V\right\| _{\ell ^{2;1}_h} \left\| \partial ^+[{\mathcal {E}}_{\mathrm {tw}}(U)]\right\| _{\ell ^\infty _h}. \end{aligned}$$
(3.34)

These are small when taking \(U = \Psi _*\) and describe the additional error contributions generated in this paper that cannot be absorbed by the terms in [28].

3.3 Initial Approximants for \({\mathcal {G}}\) and \({\mathcal {G}}^+\)

The expression (2.12) for \({\mathcal {G}}(U)\) is too convoluted for practical use, featuring third differences and double sums. It hence needs to be simplified, at the cost of introducing error terms. An initial step in this direction was performed in [28,  Eq. (6.10)], where we decomposed \({\mathcal {G}}(U)\) into a number of products featuring nonlinearities from the set

$$\begin{aligned} {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}= & {} \{ {\mathcal {Y}}_1, {\mathcal {Y}}_2, {\mathcal {D}}^{\diamond _0;+}, {\mathcal {D}}^{\diamond _-;+}, {\mathcal {X}}_A, {\mathcal {X}}_B, {\mathcal {X}}_C, {\mathcal {X}}_D \}, \end{aligned}$$
(3.35)

which were all defined in [28,  §6] and contain at most second differences. A similar decomposition was obtained for \({\mathcal {G}}^+(U) = \partial ^+ {\mathcal {G}}(U)\) in [28,  Eq. (6.16)], but now with nonlinearities from the set

$$\begin{aligned} \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}= & {} {\mathcal {S}}_{\mathrm {nl};\mathrm {short}} \cup \{ {\mathcal {Y}}_1^+, {\mathcal {Y}}_{2b}^+ \} , \end{aligned}$$
(3.36)

together with an explicit third-difference term. In addition, for each of the nonlinearities \(f \in \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}\) we (implicitly) defined an approximation \(f_{\mathrm {apx}}(U)\) and an approximate linearization \(f_{\mathrm {lin};U}[V]\) in [28,  §8].

In fact, the full definitions of the nonlinearities \(f \in \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}\) turn out to be irrelevant for our purposes here, so there is no need to repeat them from [28,  §6]. However, we do need to manipulate their approximations, which we therefore evaluate in full here by substituting the relevant expressions from [28,  §7] into the definitions [28,  Eq. (8.1)-(8.4)]. This yields the approximants

$$\begin{aligned}&\mathcal {X}_{A;\mathrm {apx}}(U) = \partial ^0 U ,\qquad \qquad \qquad \qquad \qquad \qquad {\mathcal {Y}}_{1;\mathrm {apx}}(U) = \partial ^0 U, \nonumber \\&\mathcal {X}_{B;\mathrm {apx}}(U) = S^+ [\gamma _U^{-1} ] \gamma _U^4 , \qquad \qquad \qquad \qquad {\mathcal {Y}}_{2;\mathrm {apx}}(U) = \gamma _U^{-4} \partial ^{(2)} U + g(U), \nonumber \\&\mathcal {X}_{C;\mathrm {apx}}(U) = S^+[ \gamma _U^{-1} ] (\gamma _U^4 - \gamma _U^2) , \qquad \qquad {\mathcal {Y}}^+_{1;\mathrm {apx}}(U) = \gamma _U^{-1} S^+[\partial ^{(2)} U] T^+ \gamma _U, \nonumber \\&\mathcal {X}_{D;\mathrm {apx}}(U) = S^+[ \gamma _U \partial ^0 U] \partial ^0 U , \qquad \qquad \quad ~{\mathcal {Y}}^+_{2b;\mathrm {apx}}(U) = \Big [ {\mathcal {E}}^+_{\mathrm {tw};\mathrm {apx}}(U) - \gamma _U^{-4} \partial ^+ \partial ^{(2)} U \Big ] \nonumber \\&\qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad + c_* \gamma _U^{-3} S^+[\partial ^{(2)} U], \end{aligned}$$
(3.37)

together with the approximate linearizations

$$\begin{aligned} {\mathcal {X}}_{A;\mathrm {lin};U}[V]= & {} \partial ^0 V + \partial ^0 U \big [ \sum _{-;h} {\mathcal {E}}_{\mathrm {sm}}(U) \partial ^0 V \big ] , \nonumber \\ {\mathcal {X}}_{B;\mathrm {lin};U}[V]= & {} S^+\Big [ \gamma _U^{-3} \partial ^0 U \partial ^0 V + \gamma _U^{-1} \big [ \sum _{-;h} {\mathcal {E}}_{\mathrm {sm}}(U) \partial ^0 V \big ] \Big ] \gamma _U^4 \nonumber \\&+ S^+[\gamma _U^{-1} ] ( - 4 \gamma _U^2) \partial ^0 U \partial ^0 V ,\nonumber \\ {\mathcal {X}}_{C;\mathrm {lin};U}[V]= & {} S^+\Big [ \gamma _U^{-3} \partial ^0 U \partial ^0 V + \gamma _U^{-1} \sum _{-;h} {\mathcal {E}}_{\mathrm {sm}}(U) \partial ^0 V \Big ] (\gamma _U^4 - \gamma _U^2)\nonumber \\&+ S^+[\gamma _U^{-1} ] [2 - 4 \gamma _U^2] \partial ^0 U \partial ^0 V, \nonumber \\ {\mathcal {X}}_{D;\mathrm {lin};U}[V]= & {} S^+[\gamma _U^{-1} (2 \gamma _U^2 - 1) \partial ^0 V ] \partial ^0 U + S^+[ \gamma _U \partial ^0 U ] \partial ^0 V \nonumber \\&+ S^+[ \gamma _U \partial ^0 U] \partial ^0 U \sum _{-;h} {\mathcal {E}}_{\mathrm {sm}}(U) \partial ^0 V , \end{aligned}$$
(3.38)

respectively

$$\begin{aligned} {\mathcal {Y}}_{1;\mathrm {lin};U}[V]= & {} \partial ^0 V - \partial ^0 U \big [ \sum _{-;h} {\mathcal {E}}_{\mathrm {sm}}(U) \partial ^0 V \big ] , \nonumber \\ {\mathcal {Y}}_{2;\mathrm {lin};U}[V]= & {} \gamma _U^{-2} M_U[V] + c_* \gamma _U^{-3} \partial ^0 V , \nonumber \\ {\mathcal {Y}}^+_{1;\mathrm {lin};U}[V]= & {} \big [ \gamma _U^{-3} \partial ^0 U[S^+ \partial ^{(2)} U]\partial ^0 V + \gamma _U^{-1} S^+ \partial ^{(2)} V \big ] T^+ \gamma _U \nonumber \\&- \gamma _U^{-1} S^+[\partial ^{(2)} U] T^+ \Big [ \gamma _U^{-1} \partial ^0 U \partial ^0 V + \gamma _U \sum _{-;h} {\mathcal {E}}_{\mathrm {sm}}(U) \partial ^0 V \Big ] , \nonumber \\ {\mathcal {Y}}^+_{2b;\mathrm {lin};U}[V]= & {} 4[ 6 \gamma _U^{-8} - 5 \gamma _U^{-6} ] S^+ [\partial ^{(2)} U ] T^+[ \partial ^{(2)} U] \partial ^0 V \nonumber \\&+ 4 \gamma _U^{-6} \partial ^0 U \Big [ T^+[\partial ^{(2)} U] S^+ [\partial ^{(2)} V] +S^+ [\partial ^{(2)} U] T^+ [\partial ^{(2)} V] \Big ] \nonumber \\&+ g''(U) [\partial ^0 U ] V + g'(U) \partial ^0 V . \end{aligned}$$
(3.39)

The corresponding expressions for the two remaining second-difference operators can be copied from [28,  Eq. (7.22)] and read

$$\begin{aligned} {\mathcal {D}}^{\diamond _0;+}_{\mathrm {apx}}(U)= & {} \gamma _U^{-3} S^+ [\partial ^{(2)} U] , \qquad \qquad {\mathcal {D}}^{\diamond _0;+}_{\mathrm {lin};U}[V] = 3 \gamma _U^{-5} \partial ^0 U S^+[\partial ^{(2)} U] \partial ^0 V \nonumber \\&\qquad \qquad \qquad \qquad \qquad \qquad + \gamma _U^{-3} S^+ [\partial ^{(2)} V] , \nonumber \\ {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}}(U)= & {} \quad \gamma _U^{-3} \partial ^{(2)} U , \qquad \quad \qquad {\mathcal {D}}^{\diamond _-;+}_{\mathrm {lin};U}[V] = 3 \gamma _U^{-5} \partial ^0 U [\partial ^{(2)} U] \partial ^0 V \nonumber \\&\qquad \qquad \qquad \qquad \qquad \qquad + \gamma _U^{-3} \partial ^{(2)} V . \end{aligned}$$
(3.40)

The expressions above were used in [28,  §8.1] to define an initial approximant

$$\begin{aligned} {\mathcal {G}}_{\mathrm {apx};I}(U)= & {} {\mathcal {G}}_{A;\mathrm {apx};I}(U) + {\mathcal {G}}_{B;\mathrm {apx};I}(U) + {\mathcal {G}}_{C;\mathrm {apx};I}(U) + {\mathcal {G}}_{D;\mathrm {apx};I}(U) \end{aligned}$$
(3.41)

for \({\mathcal {G}}(U)\), featuring the four components

$$\begin{aligned} {\mathcal {G}}_{A;\mathrm {apx};I}(U)= & {} \Big [ 1 - {\mathcal {Y}}_{1;\mathrm {apx}}(U) T^{-}\big [ {\mathcal {X}}_{A;\mathrm {apx}}(U) \big ] \Big ] {\mathcal {Y}}_{2;\mathrm {apx}}(U) , \nonumber \\ {\mathcal {G}}_{B;\mathrm {apx};I}(U)= & {} {\mathcal {Y}}_{1;\mathrm {apx}}(U) \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx}}(U) T^{-} \Big [ {\mathcal {X}}_{B;\mathrm {apx}}(U) \Big ] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}}(U), \nonumber \\ {\mathcal {G}}_{\#;\mathrm {apx};I}(U)= & {} {\mathcal {Y}}_{1;\mathrm {apx}}(U) \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx}}(U) T^{-} \Big [ {\mathcal {X}}_{\#;\mathrm {apx}} (U) {\mathcal {D}}^{\diamond _0;+}_{\mathrm {apx}}(U) \Big ] , \end{aligned}$$
(3.42)

for \(\# \in \{C, D\}\). In addition, we introduced the approximate linearization

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {lin};U;I}[V]= & {} {\mathcal {G}}_{A;\mathrm {lin};U;I}[V] + {\mathcal {G}}_{B;\mathrm {lin};U;I}[V] + {\mathcal {G}}_{C;\mathrm {lin};U;I}[V] + {\mathcal {G}}_{D;\mathrm {lin};U;I}[V] \end{array} \end{aligned}$$
(3.43)

by writing

$$\begin{aligned} {\mathcal {G}}_{A;\mathrm {lin};U;I}[V]= & {} - {\mathcal {Y}}_{1;\mathrm {lin};U}[V] T^{-}\big [ {\mathcal {X}}_{A;\mathrm {apx}}(U) \big ] {\mathcal {Y}}_{2;\mathrm {apx}} (U) \nonumber \\&- {\mathcal {Y}}_{1;\mathrm {apx}}(U) T^{-}\big [ {\mathcal {X}}_{A;\mathrm {lin};U}[V] \big ] {\mathcal {Y}}_{2;\mathrm {apx}} (U) \nonumber \\&+ \Big [1 - {\mathcal {Y}}_{1;\mathrm {apx}}(U) T^{-}\big [ {\mathcal {X}}_{A;\mathrm {apx}}(U) \big ] \Big ] {\mathcal {Y}}_{2;\mathrm {lin};U} [V] \end{aligned}$$
(3.44)

and applying the analogous product-rule procedure to obtain \({\mathcal {G}}_{\#; \mathrm {lin};U;I}[V]\) for \(\# \in \{B, C, D\}\); see [28,  Eq. (8.6)-(8.7)] and Sects. E.1E.3. Treating \({\mathcal {G}}^+\) in a similar spirit, we defined initial approximants

$$\begin{aligned} {\mathcal {G}}^+_{\mathrm {apx};I}(U)= & {} {\mathcal {G}}^+_{A'a;\mathrm {apx};I}(U) + {\mathcal {G}}^+_{A'b;\mathrm {apx};I}(U) + {\mathcal {G}}^+_{A'c;\mathrm {apx};I}(U) \nonumber \\&+ {\mathcal {G}}^+_{B';\mathrm {apx};I}(U) + {\mathcal {G}}^+_{C';\mathrm {apx};I}(U) + {\mathcal {G}}^+_{D';\mathrm {apx};I}(U) , \nonumber \\ {\mathcal {G}}^+_{\mathrm {lin};U;I}[V]= & {} {\mathcal {G}}^+_{A'a;\mathrm {lin};U;I}[V] + {\mathcal {G}}^+_{A'b;\mathrm {lin};U;I}[V] + {\mathcal {G}}^+_{A'c;\mathrm {lin};U;I}[V] \nonumber \\&+ {\mathcal {G}}^+_{B';\mathrm {lin};U;I}[V] + {\mathcal {G}}^+_{C';\mathrm {lin};U;I}[V] + {\mathcal {G}}^+_{D';\mathrm {lin};U;I}[V], \end{aligned}$$
(3.45)

in which we have introduced the expressions

$$\begin{aligned} {\mathcal {G}}^+_{A'a;\mathrm {apx};I}(U)= & {} \gamma _U^{-2} \partial ^+ \partial ^{(2)} U , \nonumber \\ {\mathcal {G}}^+_{A'b;\mathrm {apx};I}(U)= & {} \Big [ 1 - {\mathcal {Y}}_{1;\mathrm {apx}}(U) {\mathcal {X}}_{A;\mathrm {apx}}(U) \Big ] {\mathcal {Y}}_{2b;\mathrm {apx}}^+(U) , \nonumber \\ {\mathcal {G}}^+_{A'c;\mathrm {apx};I}(U)= & {} - {\mathcal {Y}}^+_{1;\mathrm {apx}}(U) {\mathcal {X}}_{A;\mathrm {apx}}(U) T^+ \big [ {\mathcal {Y}}_{2;\mathrm {apx}}(U) \big ] , \end{aligned}$$
(3.46)

together with

$$\begin{aligned} {\mathcal {G}}^+_{B';\mathrm {apx};I}(U)= & {} {\mathcal {Y}}^+_{1;\mathrm {apx}}(U) T^+ \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx}}(U) T^-\Big [ {\mathcal {X}}_{B;\mathrm {apx}}(U) \Big ] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}}(U) , \nonumber \\ {\mathcal {G}}^+_{\#';\mathrm {apx};I}(U)= & {} {\mathcal {Y}}^+_{1;\mathrm {apx}}(U) T^+ \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx}}(U) T^-\Big [ {\mathcal {X}}_{\#;\mathrm {apx}}(U) {\mathcal {D}}^{\diamond _0;+}_{\mathrm {apx}}(U) \Big ] , \end{aligned}$$
(3.47)

for \(\# \in \{C, D\}\). With the sole exception of

$$\begin{aligned} {\mathcal {G}}^+_{A'a;\mathrm {lin};U;I}[V]= & {} \gamma _U^2 {\widetilde{M}}_{U;III}[V] -2 \gamma _U^{-4 } \partial ^0 U [\partial ^+ \partial ^{(2)} U ] \partial ^0 V , \end{aligned}$$
(3.48)

all the approximate linearizations in (3.45) can be found by applying the product-rule procedure underpinning (3.44) to the expressions (3.46)–(3.47); see [28,  §8.2] and Sects. F.1F.3.

One of our main aims in [28] was to develop a framework to control the errors that arise by these types of approximations. In particular, we needed to track the propagation of errors on the individual factors of (2.12) through the full sums and exponents. The bounds in [28,  Lems. 8.1-8.3] provide a constant \(K=K(\kappa ) >0\) so that these errors satisfy

$$\begin{aligned} \left\| {\mathcal {G}}(U) - {\mathcal {G}}_{\mathrm {apx};I}(U) \right\|\le & {} K h , \nonumber \\ \left\| {\mathcal {G}}^+(U) - {\mathcal {G}}^+_{\mathrm {apx};I}(U) \right\| _{\ell ^2_h}\le & {} K h \big [1 + \left\| \partial ^+ \partial ^+ \partial ^+ U\right\| _{\ell ^2_h} \big ] \end{aligned}$$
(3.49)

for all \(U \in \Omega _{h;\kappa }\). In addition, the nonlinear residuals

$$\begin{aligned} {\mathcal {G}}_{\mathrm {nl};U;I}(V)= & {} {\mathcal {G}}(U + V) - {\mathcal {G}}(U) - {\mathcal {G}}_{\mathrm {lin};U;I}[V], \nonumber \\ {\mathcal {G}}^+_{\mathrm {nl};U;I}(V)= & {} {\mathcal {G}}^+(U + V) - {\mathcal {G}}^+(U) - {\mathcal {G}}^+_{\mathrm {lin};U;I}[V] \end{aligned}$$
(3.50)

can be estimated as

$$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};U;I}(V) \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {prod}}( V, V) + K {\mathcal {E}}_{\mathrm {sh};U}(V), \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {nl};U;I}(V) \right\| _{\ell ^2_h}\le & {} K \overline{{\mathcal {E}}}_{\mathrm {prod};U}( V, V) + K \overline{{\mathcal {E}}}_{\mathrm {sh};U}(V) \end{aligned}$$
(3.51)

for any \(U \in \Omega _{h;\kappa }\) and any \(V \in \ell ^2_h\) for which \(U + V \in \Omega _{h;\kappa }\).

These initial approximants for \({\mathcal {G}}\) and \({\mathcal {G}}^+\) are already much easier to work with than (2.12) and enabled us to establish the well-posedness of our reduced system (2.13) in [28]. However, they are still unwieldy on account of the shifts and the sums. In addition, several simplifications can be made that only become apparent when looking at the full combinations (3.41), (3.43) and (3.45). This will be the main focus of Sects. 45.

Convention Throughout the remainder of this paper, we use the convention that primed constants (such as \(C_1'\), \(C_2'\) etc) that appear in proofs are positive and depend only on \(\kappa \) and the nonlinearity g, unless explicitly stated otherwise.

4 Component Estimates

The first important task in this paper is to build a bridge between the linear theory described in Sect. 3.1 and the approximation framework outlined in Sect. 3.3. This requires us to refine the approximants introduced in the latter section. We carry out the first step of this procedure here, focusing our attention on the nonlinearities introduced in (3.35)–(3.36).

In particular, for any \(f \in \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}\) we introduce the further decompositions

$$\begin{aligned} f_{\mathrm {apx}}(U)= & {} f_{\mathrm {apx};\mathrm {expl}}(U) + f_{\mathrm {apx};\mathrm {sh}}(U) + f_{\mathrm {apx};\mathrm {rem}}(U) , \nonumber \\ f_{\mathrm {lin};U}[V]= & {} f_{\mathrm {lin};U;\mathrm {expl}}[V] + f_{\mathrm {lin};U;\mathrm {sh}}[V] + f_{\mathrm {lin};U;\mathrm {rem}}[V] . \end{aligned}$$
(4.1)

The expressions with the label ‘\(\mathrm {expl}\)’ are the actual explicit simplifications that will play a key role in our further computations. The label ‘\(\mathrm {sh}\)’ is used for terms which are always small, which we will be able to absorb into the error terms \({\mathcal {E}}_{\mathrm {sh}}\) and \(\overline{{\mathcal {E}}}_{\mathrm {sh}}\) defined in (3.26). Finally, the label ‘\(\mathrm {rem}\)’ is used for remainder terms that are small when using \(U = \Psi _*\).

The explicit decompositions (4.1) are provided in Sect. D. Our main contribution here is to summarize the errors that arise in a structured fashion that resembles the main spirit of the framework developed in [28,  §7]. This will allow us to replace all the occurrences of \(f_{\mathrm {apx}}\) and \(f_{\mathrm {lin};U}\) in Sect. 3.3 by their refinements \(f_{\mathrm {apx};\mathrm {expl}}\) and \(f_{\mathrm {lin};U;\mathrm {expl}}\), leading to a second round of approximations \({\mathcal {G}}_{\mathrm {apx};II}\), \({\mathcal {G}}^+_{\mathrm {apx};II}\), \({\mathcal {G}}_{\mathrm {lin};U;II}\) and \({\mathcal {G}}^+_{\mathrm {lin};U;II}\).

In order to achieve this, we define a preferred exponent set

$$\begin{aligned} Q_{f;\mathrm {pref}} \subset \{2, \infty \} \end{aligned}$$
(4.2)

for each \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\), together with its counterpart

$$\begin{aligned} {\overline{Q}}_{f;\mathrm {pref}} \subset \{2, \infty \} \end{aligned}$$
(4.3)

for each \(f \in \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}\). This is done in such a way that we can write

$$\begin{aligned} {\mathcal {G}}_{\mathrm {apx};I}(U)= & {} \sum _{i=1}^N \pi _i \big [ f_{\mathrm {apx};i;1}(U) , \ldots , f_{\mathrm {apx};i; k_i}(U) \big ], \nonumber \\ {\mathcal {G}}^+_{\mathrm {apx};I}(U) - {\mathcal {G}}^+_{A'a;\mathrm {apx};I}(U)= & {} \sum _{i=1}^N {\overline{\pi }}_i \big [ {\overline{f}}_{\mathrm {apx};i;1}(U) , \ldots , {\overline{f}}_{\mathrm {apx};i; k_i}(U) \big ] , \end{aligned}$$
(4.4)

for a set of bounded multi-linear maps

$$\begin{aligned} \pi _i: \ell ^{q_{i;1}}_h \times \ldots \times \ell ^{q_{i;k_i}}_h \rightarrow \ell ^2_h, \qquad \qquad {\overline{\pi }}_i: \ell ^{{\overline{q}}_{i;1}}_h \times \ldots \times \ell ^{{\overline{q}}_{i;k_i}}_h \rightarrow \ell ^2_h, \end{aligned}$$
(4.5)

each defined for \(1 \le i \le N\), where we have the inclusions

$$\begin{aligned} f_{i;j} \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}, \qquad q_{i; j} \in Q_{f_{i;j};\mathrm {pref}}, \qquad \qquad {\overline{f}}_{i;j} \in \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}, \qquad {\overline{q}}_{i; j} \in {\overline{Q}}_{{\overline{f}}_{i;j};\mathrm {pref}}, \end{aligned}$$
(4.6)

for all \(1 \le j \le k_i\). Stated more informally, the \(\ell ^2_h\) norm of \({\mathcal {G}}_{\mathrm {apx};I}(U)\) can be bounded in terms of products of \(\ell ^q_2\)-norms of nonlinearities \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\), where each q is taken from the preferred set of exponents. This is the direct analogue of [28,  Cor. 6.4].

A short inspection of the products (3.42), (3.46) and (3.47) readily shows that there is some freedom as to which factors should be measured in \(\ell ^2_h\). In fact, it is possible to put an \(\ell ^2_h\) norm on any chosen factor, at the price of possibly having to flip the exponent of a companion factor that has \(2 \in Q_{f; \mathrm {pref}}\) from two to infinity.

This freedom is essential to obtain sharp estimates and hence requires us to deviate from the preferred exponents from time to time. The main focus of [28,  §5,§7-8] was to develop a bookkeeping framework to keep track of this procedure. We build on this investment here and follow the spirit of [28,  §5.2] to define further exponent sets

$$\begin{aligned} Q_f \subset \{2, \infty \}, \qquad Q_{f;\mathrm {lin}} \subset \{2, \infty \}, \qquad Q_{f;\mathrm {lin};\mathrm {rem}}\subset \{2, \infty \} \end{aligned}$$
(4.7)

for each \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}} \cup \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}\). The first of these contains all values of q for which \(f_{\mathrm {apx}}\) maps into \(\ell ^q_h\). On the other hand, the set \(Q_{f;\mathrm {lin}}\) contains all q for which we need to evaluate the \(\ell ^q_h\)-norm of \(f_{\mathrm {lin};U;\mathrm {expl}}\) and \(f_{\mathrm {lin};U;\mathrm {sh}}\), while \(Q_{f;\mathrm {lin};\mathrm {rem}}\) contains these exponents for \(f_{\mathrm {lin};U;\mathrm {rem}}\).

In order to illustrate these points, let us consider the example

$$\begin{aligned} \begin{array}{lcl} {\mathcal {I}}_{\mathrm {ex};I;U}[V]= & {} \sum _{-;h}{\mathcal {Y}}_{2;\mathrm {apx}}(U) {\mathcal {X}}_{B;\mathrm {lin};U}[V] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U), \end{array} \end{aligned}$$
(4.8)

which appears (after dropping a shift for notational clarity) as a factor in the component \({\mathcal {G}}_{B;\mathrm {lin};U;I}\) that needs to be evaluated in the supremum norm; see (E.17). Our goal is to simplify this expression by writing

$$\begin{aligned} \begin{array}{lcl} {\mathcal {I}}_{\mathrm {ex};II;U}[V]= & {} \sum _{-;h}{\mathcal {Y}}_{2;\mathrm {apx};\mathrm {expl}}(U) {\mathcal {X}}_{B;\mathrm {lin};U;\mathrm {expl}}[V] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U), \end{array} \end{aligned}$$
(4.9)

noting that \({\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}}\) is not simplified further; see Sect. D. Exploiting the fact that \({\mathcal {Y}}_{2;\mathrm {apx};\mathrm {sh}} = 0\), a short computation readily yields the decomposition

$$\begin{aligned} {\mathcal {I}}_{\mathrm {ex};I;U}[V] = {\mathcal {I}}_{\mathrm {ex};II;U}[V] + {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};a} + {\mathcal {I}}_{\mathrm {ex};\mathrm {sh};a} + {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};b}, \end{aligned}$$
(4.10)

where we have introduced the terms

$$\begin{aligned} {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};a}= & {} \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx}}(U) {\mathcal {X}}_{B;\mathrm {lin};\mathrm {rem};U}[V] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U), \nonumber \\ {\mathcal {I}}_{\mathrm {ex};\mathrm {sh};a}= & {} \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx}}(U) {\mathcal {X}}_{B;\mathrm {lin};U;\mathrm {sh}}[V] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U), \nonumber \\ {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};b}= & {} \sum _{-;h} {\mathcal {Y}}_{2;\mathrm {apx};\mathrm {rem}}(U) {\mathcal {X}}_{B;\mathrm {lin};U;\mathrm {expl}}[V] {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U). \end{aligned}$$
(4.11)

We note here that the preferred exponent sets are defined in (D.12), (D.19) and (D.30) and given by

$$\begin{aligned} Q_{{\mathcal {Y}}_{2};\mathrm {pref}} = \{ 2\}, \qquad Q_{{\mathcal {X}}_{B};\mathrm {pref}} = \{ \infty \}, \qquad Q_{{\mathcal {D}}^{\diamond _-;+};\mathrm {pref}} = \{ 2\}. \end{aligned}$$
(4.12)

Recalling the remainder function introduced in (3.34), we may readily use these preferred exponents to compute

$$\begin{aligned} \left\| {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};a}\right\| _{\ell ^\infty _h} \le \left\| {\mathcal {Y}}_{2;\mathrm {apx}}(U)\right\| _{\ell ^2_h} \left\| {\mathcal {X}}_{B;\mathrm {lin};\mathrm {rem};U}[V]\right\| _{\ell ^\infty _h} \left\| {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U)\right\| _{\ell ^2_h} \le K {\mathcal {E}}_{\mathrm {rem};U}(V).\nonumber \\ \end{aligned}$$
(4.13)

Here we use property (4.33) below with \(q =\infty \) (see (D.36)) together with the a-priori bounds (4.21). For the second term we can use the same properties, but now with \(q = 2\) (see (D.36)). In particular, we obtain

$$\begin{aligned} \left\| {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};a}\right\| _{\ell ^\infty _h} \le \left\| {\mathcal {Y}}_{2;\mathrm {apx}}(U)\right\| _{\ell ^\infty _h} \left\| {\mathcal {X}}_{B;\mathrm {lin};\mathrm {sh};U}[V]\right\| _{\ell ^2_h} \left\| {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U)\right\| _{\ell ^2_h} \le K {\mathcal {E}}_{\mathrm {sh};U}(V).\nonumber \\ \end{aligned}$$
(4.14)

Note that this required us to swap the first two exponents, which is made possible by the demand \(\infty \in Q_f\) for each \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\); see Proposition 4.1.

This swap is also required for the final term, which can be controlled by

$$\begin{aligned} \left\| {\mathcal {I}}_{\mathrm {ex};\mathrm {rem};b}\right\| _{\ell ^\infty _h} \le \left\| {\mathcal {Y}}_{2;\mathrm {apx};\mathrm {rem}}(U)\right\| _{\ell ^\infty _h} \left\| {\mathcal {X}}_{B;\mathrm {lin};U;\mathrm {expl}}[V]\right\| _{\ell ^2_h} \left\| {\mathcal {D}}^{\diamond _-;+}_{\mathrm {apx}} (U)\right\| _{\ell ^2_h} \le K {\mathcal {E}}_{\mathrm {rem};U}(V).\nonumber \\ \end{aligned}$$
(4.15)

Indeed, simply using \(\ell ^\infty _h\) on the middle factor would lead to a contribution proportional to \(\left\| V\right\| _{\ell ^\infty _h}\); see the third line of (D.32). Such a term would lead to problems in Sect. 6 and hence is not contained in \({\mathcal {E}}_{\mathrm {rem};U}\). This scenario is covered in our structural results below by using options (b) from both Proposition 4.2 and 4.3. We feel that this relatively small example already clearly illustrates the benefits of utilizing an abstract bookkeeping scheme instead of direct estimates.

4.1 Summary of Estimates

In order to state our results, we introduce the expressions

$$\begin{aligned} \begin{array}{lclclcl} S_{\mathrm {sh};\mathrm {full}}(U) &{} = &{} h , &{} \qquad \qquad &{} S_{\mathrm {sh};2;\mathrm {fix}}(U) &{} = &{} 0 , \\ {\overline{S}}_{\mathrm {sh};\mathrm {full}}(U) &{} = &{} h \big [ 1 + \left\| \partial ^+\partial ^+\partial ^+ U\right\| _{\ell ^2_h} + \left\| \partial ^+\partial ^+\partial ^+ U\right\| _{\ell ^\infty _h} \big ] , &{} \qquad \qquad &{} {\overline{S}}_{\mathrm {sh};2;\mathrm {fix}}(U) &{} = &{} 0 , \\ \end{array}\nonumber \\ \end{aligned}$$
(4.16)

together with

$$\begin{aligned} \begin{array}{lclclcl} S_{\mathrm {rem};\mathrm {full}}(U) &{} = &{} \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^2_h} + \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^\infty _h} , &{} \qquad \qquad &{} S_{\mathrm {rem};2;\mathrm {fix}}(U) &{} = &{} 0 , \\ {\overline{S}}_{\mathrm {rem};\mathrm {full}}(U) &{} = &{} S_{\mathrm {rem};\mathrm {full}}(U) + \left\| \partial ^+[ {\mathcal {E}}_{\mathrm {tw}}(U) ]\right\| _{\ell ^2_h} , &{} \qquad \qquad &{} {\overline{S}}_{\mathrm {rem};2;\mathrm {fix}}(U) &{} = &{} 0 \\ \end{array} \end{aligned}$$
(4.17)

and finally

$$\begin{aligned} \begin{array}{lcl} S_{\mathrm {diff};\mathrm {full}}(U^{(1)}, U^{(2)}) &{} = &{} \left\| U^{(2)} - U^{(1)}\right\| _{\ell ^{2;2}_h} + \left\| U^{(2)} - U^{(1)}\right\| _{\ell ^{\infty ;1}_h} , \\ S_{\mathrm {diff};2;\mathrm {fix}}(U^{(1)}, U^{(2)}) &{} = &{} \left\| U^{(2)} - U^{(1)}\right\| _{\ell ^{2;2}_h}. \\ \end{array} \end{aligned}$$
(4.18)

These expressions are all related to the size of the \(f_{\mathrm {apx}}\) functions and play a very similar role as the quantities \(S_{\mathrm {full}}\) and \(S_{2;\mathrm {fix}}\) that were defined in [28,  §7]. In particular, the ‘full’ terms correspond to all the exponents that we need to use, while the ‘fix’ expressions reflect the contributions that are only allowed to be evaluated in \(\ell ^2_h\); see (4.26).

In addition, we recall the quantities

$$\begin{aligned} \begin{array}{lclclcl} T_{\mathrm {safe}}(V) &{} = &{} \left\| V\right\| _{\ell ^{2;2}_h} , &{} \qquad \qquad &{} {\overline{T}}_{\mathrm {safe}}(V) &{} = &{} T_{\mathrm {safe}}(V) , \\ T_{\infty ;\mathrm {opt}}(V) &{} = &{} \left\| \partial ^+ V\right\| _{\ell ^\infty _h} , &{} &{} {\overline{T}}_{\infty ;\mathrm {opt}}(V) &{} = &{} T_{\infty ;\mathrm {opt}}(V) + \left\| \partial ^+ \partial ^+ V\right\| _{\ell ^\infty _h} \end{array} \end{aligned}$$
(4.19)

that are associated to the approximate linearizations \(f_{\mathrm {lin}}\). Here \(T_{\infty ;\mathrm {opt}}\) represents the contributions where the use of the supremum norm is optional, in the sense that they could also be measured in \(\ell ^2_h\). The remaining contributions are all reflected in \(T_{\mathrm {safe}}\). We emphasize that the main point of our bookkeeping scheme is to ensure that products of the form \(S_{\#;\mathrm {full}} T_{\infty ;\mathrm {opt}}\) are never needed, where \(\# \in \{ \mathrm {sh},\mathrm {rem},\mathrm {diff} \}\).

Our main results summarize the structure that the decompositions described in Sect. D will adhere to. Propositions 4.1 and 4.2 state that the approximants \(f_{\mathrm {apx};\#}\) are all uniformly bounded and that the full linear approximants \(f_{\mathrm {lin};U}\) share the structure and estimates of the nonlinearities in the sets \({\mathcal {S}}_{\mathrm {nl}} \cup \overline{{\mathcal {S}}}_{\mathrm {nl}}\) analyzed in [28]. These can be interpreted as the counterparts of [28,  Cor. 7.6 and 7.8]. On the other hand, Propositions 4.3-4.4 should be seen as the equivalents of [28,  Cor. 7.7], while Propositions 4.5-4.6 are the equivalents of [28,  Cor. 7.9].

Proposition 4.1

(see Sect. D) For every \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\) we have \(\infty \in Q_f\) together with

$$\begin{aligned} Q_{f;\mathrm {pref}} \subset Q_f \cap Q_{f;\mathrm {lin}} \cap Q_{f;\mathrm {lin};\mathrm {rem}}. \end{aligned}$$
(4.20)

In addition, there exists \(K > 0\) so that for each \(q \in Q_f\), the bound

$$\begin{aligned} \left\| f_{\mathrm {apx};\mathrm {expl}}(U)\right\| _{\ell ^q_h} + \left\| f_{\mathrm {apx};\mathrm {sh}}(U)\right\| _{\ell ^q_h} + \left\| f_{\mathrm {apx};\mathrm {rem}}(U)\right\| _{\ell ^q_h} \le K \end{aligned}$$
(4.21)

holds for all \(h > 0\) and \(U \in \Omega _{h;\kappa }\). The same properties hold upon replacing \(({\mathcal {S}}_{\mathrm {nl};\mathrm {short}}, Q_{f;\mathrm {pref}})\) by \((\overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}, {\overline{Q}}_{f;\mathrm {pref}})\).

Proposition 4.2

(see Sect. D) Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). For any \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\), any \(\# \in \{\mathrm {expl},\mathrm {sh},\mathrm {rem}\}\) and any \(q \in Q_{f;\mathrm {pref}}\), at least one of the following two properties hold true.

  1. (a)

    There exists \(K > 0\) so that

    $$\begin{aligned} \left\| f_{\mathrm {lin};U;\#}[V]\right\| _{\ell ^q_h} \le K T_{\mathrm {safe}}(V) \end{aligned}$$
    (4.22)

    holds for every \(h> 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\).

  2. (b)

    We have \(q = \infty \) and there exists \(K > 0\) so that the bounds

    $$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {lin};U;\#}[V]\right\| _{\ell ^2_h} &{} \le &{} K T_{\mathrm {safe}}(V), \\ \left\| f_{\mathrm {lin};U;\#}[V]\right\| _{\ell ^\infty _h} &{} \le &{} K T_{\infty ;\mathrm {opt}}(V) \end{array} \end{aligned}$$
    (4.23)

    hold for every \(h> 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\).

The same properties hold upon making the replacement

$$\begin{aligned} ({\mathcal {S}}_{\mathrm {nl};\mathrm {short}}, Q_{f;\mathrm {pref}}, T_{\mathrm {safe}}, T_{\infty ;\mathrm {opt}}) \mapsto (\overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}, {\overline{Q}}_{f;\mathrm {pref}}, {\overline{T}}_{\mathrm {safe}}, {\overline{T}}_{\infty ;\mathrm {opt}}). \end{aligned}$$
(4.24)

Proposition 4.3

(see Sect. D) Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). Then there exists \(K > 0\) so that for every \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\), \(q \in Q_{f;\mathrm {pref}}\) and \(\# \in \{\mathrm {sh}, \mathrm {rem}\}\), we have

$$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {apx};\#}(U)\right\| _{\ell ^q_h}\le & {} K S_{\#;\mathrm {full}}(U) \end{array} \end{aligned}$$
(4.25)

for any \(h > 0\) and \(U \in \Omega _{h;\kappa }\).

In addition, if \(2 \in Q_{f;\mathrm {pref}}\) then for every \(\# \in \{\mathrm {sh}, \mathrm {rem}\}\) at least one of the following two properties hold true.

  1. (a)

    There exists \(K > 0\) so that

    $$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {apx};\#}(U)\right\| _{\ell ^2_h}\le & {} K S_{\#;2;\mathrm {fix}}(U) \end{array} \end{aligned}$$
    (4.26)

    holds for every \(h > 0\) and \(U \in \Omega _{h;\kappa }\).

  2. (b)

    There exists \(K>0\) so that

    $$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {apx};\#}(U)\right\| _{\ell ^\infty _h} &{} \le &{} K S_{\#;\mathrm {full}}(U) \\ \end{array} \end{aligned}$$
    (4.27)

    holds for every \(h > 0\) and \(U \in \Omega _{h;\kappa }\).

The same properties hold upon making the replacement

$$\begin{aligned} ({\mathcal {S}}_{\mathrm {nl};\mathrm {short}}, Q_{f;\mathrm {pref}}, S_{\#;\mathrm {full}}, S_{\#;2;\mathrm {fix}}) \mapsto (\overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}, {\overline{Q}}_{f;\mathrm {pref}}, {\overline{S}}_{\#;\mathrm {full}}, {\overline{S}}_{\#;2;\mathrm {fix}}). \end{aligned}$$
(4.28)

Proposition 4.4

(see Sect. D) Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). Then there exists \(K > 0\) so that for every \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\), \(q \in Q_{f;\mathrm {pref}}\) and \(\# \in \{\mathrm {expl},\mathrm {sh}, \mathrm {rem}\}\), we have

$$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {apx};\#}(U^{(2)} ) - f_{\mathrm {apx};\#}(U^{(1)} ) \right\| _{\ell ^q_h}\le & {} K S_{\mathrm {diff};\mathrm {full}}(U^{(1)} , U^{(2)}) \end{array} \end{aligned}$$
(4.29)

for any \(h > 0\) and any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\).

In addition, if \(2 \in Q_{f;\mathrm {pref}}\), then for every \(\# \in \{\mathrm {expl},\mathrm {sh}, \mathrm {rem}\}\) at least one of the following two properties hold true.

  1. (a)

    There exists \(K > 0\) so that

    $$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {apx};\#}(U^{(2)} ) - f_{\mathrm {apx};\#}(U^{(1)} ) \right\| _{\ell ^2_h}\le & {} K S_{\mathrm {diff};2;\mathrm {fix}}(U^{(1)} , U^{(2)}) \end{array} \end{aligned}$$
    (4.30)

    holds for every \(h > 0\) and any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\).

  2. (b)

    There exists \(K>0\) so that

    $$\begin{aligned} \begin{array}{lcl} \left\| f_{\mathrm {apx};\#}(U^{(2)} ) - f_{\mathrm {apx};\#}(U^{(1)} ) \right\| _{\ell ^\infty _h} &{} \le &{} K S_{\mathrm {diff};\mathrm {full}}(U^{(1)} , U^{(2)}) \\ \end{array} \end{aligned}$$
    (4.31)

    holds for every \(h > 0\) and any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\).

Proposition 4.5

(see Sect. D) Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). Consider any \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\) and any \(\# \in \{\mathrm {sh}, \mathrm {rem}\}\). Then if \(2 \in Q_{f;\mathrm {pref}}\), there exists a constant \(K > 0 \) so that

$$\begin{aligned} \left\| f_{\mathrm {lin};U;\#}(V)\right\| _{\ell ^2_h} \le K {\mathcal {E}}_{\#;U}(V) \end{aligned}$$
(4.32)

holds for all \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) for which \(U + V \in \Omega _{h;\kappa }\).

Otherwise, there exists \(q \in \{2, \infty \}\) together with a constant \(K > 0\) so that

$$\begin{aligned} \left\| f_{\mathrm {lin};U;\#}(V)\right\| _{\ell ^q_h} \le K {\mathcal {E}}_{\#;U}(V) \end{aligned}$$
(4.33)

holds for all \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) for which \(U + V \in \Omega _{h;\kappa }\). The same properties hold upon making the replacement

$$\begin{aligned} ({\mathcal {S}}_{\mathrm {nl};\mathrm {short}}, Q_{f;\mathrm {pref}}, {\mathcal {E}}_{\#}) \mapsto (\overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}, {\overline{Q}}_{f;\mathrm {pref}}, \overline{{\mathcal {E}}}_{\#}). \end{aligned}$$
(4.34)

Proposition 4.6

(see Sect. D)

Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). Consider any \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\) and any \(\# \in \{\mathrm {expl},\mathrm {sh}, \mathrm {rem}\}\). Then if \(2 \in Q_{f;\mathrm {pref}}\), there exists a constant \(K > 0 \) so that

$$\begin{aligned} \left\| f_{\mathrm {lin};U^{(2)};\#}(V) - f_{\mathrm {lin};U^{(1)};\#}(V) \right\| _{\ell ^2_h} \le K {\mathcal {E}}_{\mathrm {prod}}(U^{(2)} - U^{(1)} ,V) \end{aligned}$$
(4.35)

holds for all \(h > 0\), any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\) and any \(V \in \ell ^2_h\).

Otherwise, there exists \(q \in \{2, \infty \}\) together with a constant \(K > 0\) so that

$$\begin{aligned} \left\| f_{\mathrm {lin};U^{(2)};\#}(V) - f_{\mathrm {lin};U^{(1)};\#}(V) \right\| _{\ell ^q_h} \le K {\mathcal {E}}_{\mathrm {prod}}(U^{(2)} - U^{(1)} ,V) \end{aligned}$$
(4.36)

holds for all \(h > 0\), any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\) and any \(V \in \ell ^2_h\).

4.2 Refined Approximants for \({\mathcal {G}}\) and \({\mathcal {G}}^+\)

We now introduce the expressions

$$\begin{aligned} \{{\mathcal {G}}_{\#;\mathrm {apx};II}(U), \, {\mathcal {G}}_{\#;\mathrm {lin};U;II}(U), \, {\mathcal {G}}^+_{\#';\mathrm {apx};II}(U), \, {\mathcal {G}}^+_{\#';\mathrm {lin};U;II}(U) \} \end{aligned}$$
(4.37)

for \(\# \in \{A, B, C, D\}\) respectively \(\#' \in \{A'b, A'c, B', C', D'\}\) by inspecting the definitions of their predecessors labelled by I in Sect. 3.3 (see (3.42), (3.43) and (3.46)) and making the replacements

$$\begin{aligned} f_{\mathrm {apx}}(U) \mapsto f_{\mathrm {apx};\mathrm {expl}}(U), \qquad f_{\mathrm {lin};U}[V] \mapsto f_{\mathrm {lin};U;\mathrm {expl}}[V] \end{aligned}$$
(4.38)

for each \(f \in {\mathcal {S}}_{\mathrm {nl};\mathrm {short}}\cup \overline{{\mathcal {S}}}_{\mathrm {nl};\mathrm {short}}\). The full explicit forms can be found in Sects. EF, but are not important for our purposes here. We leave the expressions for \(A'a\) intact and simply write

$$\begin{aligned} {\mathcal {G}}^+_{A'a;\mathrm {apx};II}(U) = {\mathcal {G}}^+_{A'a;\mathrm {apx};I}(U), \qquad \qquad {\mathcal {G}}^+_{A'a;\mathrm {lin};U;II}[V] = {\mathcal {G}}^+_{A'a;\mathrm {apx};U;II}[V]. \end{aligned}$$
(4.39)

Our interest here is in the refined approximants

$$\begin{aligned} {\mathcal {G}}_{\mathrm {apx};II}(U)= & {} {\mathcal {G}}_{A;\mathrm {apx};II}(U) + {\mathcal {G}}_{B;\mathrm {apx};II}(U) \nonumber \\&+ {\mathcal {G}}_{C;\mathrm {apx};II}(U) + {\mathcal {G}}_{D;\mathrm {apx};II}(U)\nonumber \\ {\mathcal {G}}^+_{\mathrm {apx};II}(U)= & {} {\mathcal {G}}^+_{A'a;\mathrm {apx};II}(U) + {\mathcal {G}}^+_{A'b;\mathrm {apx};II}(U) \nonumber \\&+ {\mathcal {G}}^+_{A'c;\mathrm {apx};II}(U)\nonumber \\&+ {\mathcal {G}}^+_{B';\mathrm {apx};II}(U) + {\mathcal {G}}^+_{C';\mathrm {apx};II}(U) \nonumber \\&+ {\mathcal {G}}^+_{D';\mathrm {apx};II}(U) \end{aligned}$$
(4.40)

and the corresponding linearizations

$$\begin{aligned} {\mathcal {G}}_{\mathrm {lin};U;II}[V]= & {} {\mathcal {G}}_{A;\mathrm {lin};U;II}[V] + {\mathcal {G}}_{B;\mathrm {lin};U;II}[V] \nonumber \\&+ {\mathcal {G}}_{C;\mathrm {lin};U;II}[V] + {\mathcal {G}}_{D;\mathrm {lin};U;II}[V]\nonumber \\ {\mathcal {G}}^+_{\mathrm {lin};U;II}[V]= & {} {\mathcal {G}}^+_{A'a;\mathrm {lin};U;II}[V] + {\mathcal {G}}^+_{A'b;\mathrm {lin};U;II}[V] \nonumber \\&+ {\mathcal {G}}^+_{A'c;\mathrm {lin};U;II}[V]\nonumber \\&+ {\mathcal {G}}^+_{B';\mathrm {lin};U;II}[V] + {\mathcal {G}}^+_{C';\mathrm {lin};U;II}[V] + {\mathcal {G}}^+_{D';\mathrm {lin};U;II}[V] . \end{aligned}$$
(4.41)

In particular, the results below describe the residuals that arise when replacing the initial approximants defined in (3.41), (3.43) and (3.45) by these refined versions. Since our bookkeeping framework has the same overall structure as in [28], we can reuse the analysis developed there in a streamlined fashion.

Lemma 4.7

Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). There exists a constant \(K > 0\) together with sequences

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {apx};\mathrm {sh};a}(U) \in \ell ^2_h, \qquad {\mathcal {G}}_{\mathrm {apx};\mathrm {rem};a}(U) \in \ell ^2_h, \qquad {\mathcal {G}}^+_{\mathrm {apx};\mathrm {sh};a}(U) \in \ell ^2_h, \qquad {\mathcal {G}}^+_{\mathrm {apx};\mathrm {rem};a}(U) \in \ell ^2_h, \end{array}\nonumber \\ \end{aligned}$$
(4.42)

defined for every \(h > 0\) and \(U \in \Omega _{h;\kappa }\), so that the following properties hold true.

  1. (i)

    For every \(h > 0\) and \(U \in \Omega _{h;\kappa }\) we have the identities

    $$\begin{aligned} {\mathcal {G}}_{\mathrm {apx};I}(U)= & {} {\mathcal {G}}_{\mathrm {apx};II}(U) + {\mathcal {G}}_{\mathrm {apx};\mathrm {sh};a}(U) + {\mathcal {G}}_{\mathrm {apx};\mathrm {rem};a}(U), \nonumber \\ {\mathcal {G}}^+_{\mathrm {apx};I}(U)= & {} {\mathcal {G}}^+_{\mathrm {apx};II}(U) + {\mathcal {G}}^+_{\mathrm {apx};\mathrm {sh};a}(U) + {\mathcal {G}}^+_{\mathrm {apx};\mathrm {rem};a}(U). \end{aligned}$$
    (4.43)
  2. (ii)

    For every \(h > 0\) and \(U \in \Omega _{h;\kappa }\) we have the bounds

    $$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {apx};\mathrm {sh};a}(U) \right\| _{\ell ^2_h}\le & {} K S_{\mathrm {sh};\mathrm {full}}(U),\nonumber \\ \left\| {\mathcal {G}}_{\mathrm {apx};\mathrm {rem};a}(U) \right\| _{\ell ^2_h}\le & {} K S_{\mathrm {rem};\mathrm {full}}(U) , \end{aligned}$$
    (4.44)

    together with

    $$\begin{aligned} \left\| {\mathcal {G}}^+_{\mathrm {apx};\mathrm {sh};a}(U) \right\| _{\ell ^2_h}\le & {} K {\overline{S}}_{\mathrm {sh};\mathrm {full}}(U) , \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {apx};\mathrm {rem};a}(U) \right\| _{\ell ^2_h}\le & {} K {\overline{S}}_{\mathrm {rem};\mathrm {full}}(U) . \end{aligned}$$
    (4.45)

Proof

Restricting ourselves to \({\mathcal {G}}\), we consider a single term of the sum (4.4). Dropping the index i, we introduce the corresponding expression

$$\begin{aligned} \begin{array}{lcl} {\mathcal {I}}_\pi (U)= & {} \pi \big [ f_{1;\mathrm {apx}}(U), \ldots ,f_{k;\mathrm {apx}}(U) \big ] - \pi \big [f_{1;\mathrm {apx};\mathrm {expl}}(U) , \ldots , f_{k;\mathrm {apx};\mathrm {expl}}(U) \big ]. \end{array} \end{aligned}$$
(4.46)

Recalling the general identity

$$\begin{aligned} (a_1 + b_1)(a_2 + b_2) (a_3 + b_3) - a_1 a_2 a_3 = b_1(a_2 + b_2) (a_3 + b_3) + a_1 b_2 (a_3 + b_3) + a_1 a_2 b_3\nonumber \\ \end{aligned}$$
(4.47)

and its extensions, we write

$$\begin{aligned} {\mathcal {I}}_{\pi ;\#}(U)= & {} \pi \big [ f_{1;\mathrm {apx};\#}(U) , f_{2;\mathrm {apx}}(U) , \ldots , f_{k;\mathrm {apx}}(U) \big ] \nonumber \\&+ \pi \big [ f_{1;\mathrm {apx};\mathrm {expl}}(U) , f_{2;\mathrm {apx};\#}(U) , \ldots , f_{k;\mathrm {apx}}(U) \big ] \nonumber \\&+ \ldots + \pi \big [ f_{1;\mathrm {apx};\mathrm {expl}}(U) , f_{2;\mathrm {apx};\mathrm {expl}}(U) , \ldots , f_{k;\mathrm {apx};\#}(U) \big ] \end{aligned}$$
(4.48)

for \(\# \in \{\mathrm {sh}, \mathrm {rem}\}\) and observe that

$$\begin{aligned} \begin{array}{lcl} {\mathcal {I}}_\pi (U)= & {} {\mathcal {I}}_{\pi ;\mathrm {sh}}(U) + {\mathcal {I}}_{\pi ;\mathrm {rem}}(U). \end{array} \end{aligned}$$
(4.49)

We now use (4.21) together with Proposition 4.3 to derive the bound

$$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {I}}_{\pi ;\# }(U) \right\| _{\ell ^2_h}\le & {} C_1' S_{\#;\mathrm {full}}(U). \end{array} \end{aligned}$$
(4.50)

The desired estimates now follow from the fact that \({\mathcal {G}}_{\mathrm {apx};I}(U) - {\mathcal {G}}_{\mathrm {apx};II}(U)\) can be written as a sum of expressions of the form (4.46). \(\square \)

Lemma 4.8

Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). There exists a constant \(K > 0\) together with linear maps

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};a} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ) , \qquad \qquad {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};a} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ) \end{array} \end{aligned}$$
(4.51)

and their counterparts

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};a} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ), \qquad \qquad {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};a} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ), \end{array} \end{aligned}$$
(4.52)

defined for all \(h > 0\) and \(U \in \Omega _{h;\kappa }\), so that the following properties hold true.

  1. (i)

    For every \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) we have the identities

    $$\begin{aligned} {\mathcal {G}}_{\mathrm {lin};U;I}[V]= & {} {\mathcal {G}}_{\mathrm {lin};U;II}[V] + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};a}[V] + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};a}[V] , \nonumber \\ {\mathcal {G}}^+_{\mathrm {lin};U;I}[V]= & {} {\mathcal {G}}^+_{\mathrm {lin};U;II}[V] + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};a}[V] + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};a}[V] . \end{aligned}$$
    (4.53)
  2. (ii)

    For every \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) we have the bounds

    $$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};a}[V] \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {sh};U}(V) , \nonumber \\ \left\| {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};a}[V] \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {rem};U}(V) , \end{aligned}$$
    (4.54)

    together with

    $$\begin{aligned} \left\| {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};a}[V] \right\| _{\ell ^2_h}\le & {} K \overline{{\mathcal {E}}}_{\mathrm {sh};U}(V) , \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};a}[V] \right\| _{\ell ^2_h}\le & {} K \overline{{\mathcal {E}}}_{\mathrm {rem};U}(V) . \end{aligned}$$
    (4.55)
  3. (iii)

    For every \(h > 0\), any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\) and any \(V \in \ell ^2_h\), we have the bound

    $$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {G}}_{\mathrm {lin};U^{(2)};\mathrm {rem};a}[V] - {\mathcal {G}}_{\mathrm {lin};U^{(1)};\mathrm {rem};a}[V] \right\| _{\ell ^2_h} &{} \le &{} K {\mathcal {E}}_{\mathrm {prod}}(U^{(2)} - U^{(1)} , V) . \\ \end{array} \end{aligned}$$
    (4.56)

Proof

Restricting ourselves to \({\mathcal {G}}\), we again consider a single term of the sum (4.4). Dropping the index i, we introduce the two corresponding expressions

$$\begin{aligned} {\mathcal {I}}_{\pi ;a;U}[V]= & {} \pi \big [ f_{1;\mathrm {lin};U}[V], f_{2;\mathrm {apx}}(U), \ldots ,f_{k;\mathrm {apx}}(U) \big ] \nonumber \\&- \pi \big [f_{1;\mathrm {lin};U}[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U), \ldots , f_{k;\mathrm {apx};\mathrm {expl}}(U) \big ] , \nonumber \\ {\mathcal {I}}_{\pi ;b;U}[V]= & {} \pi \big [f_{1;\mathrm {lin};U}[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U), \ldots , f_{k;\mathrm {apx};\mathrm {expl}}(U) \big ] \nonumber \\&- \pi \big [f_{1;\mathrm {lin};U;\mathrm {expl}}[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U), \ldots , f_{k;\mathrm {apx};\mathrm {expl}}(U) \big ]. \end{aligned}$$
(4.57)

Writing

$$\begin{aligned} {\mathcal {I}}_{\pi ;a;U;\#}[V]= & {} \pi \Big [ f_{1;\mathrm {lin};U}[V], f_{2;\mathrm {apx};\#}(U), \ldots , f_{k;\mathrm {apx}}(U) \Big ] \nonumber \\&+ \ldots + \pi \Big [ f_{1;\mathrm {lin};U}[V], f_{2;\mathrm {apx};\mathrm {expl}}(U), \ldots , f_{k;\mathrm {apx};\#}(U) \Big ],\nonumber \\ {\mathcal {I}}_{\pi ;b;U;\#}[V]= & {} \pi \big [ f_{1;\mathrm {lin};U;\#}[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U), \ldots , f_{k;\mathrm {apx};\mathrm {expl}}(U) \big ] \end{aligned}$$
(4.58)

for \(\# \in \{\mathrm {sh}, \mathrm {rem}\}\), we see that

$$\begin{aligned} {\mathcal {I}}_{\pi ;a;U}[V]= & {} {\mathcal {I}}_{\pi ;a;U;\mathrm {sh}}[V] + {\mathcal {I}}_{\pi ;a;U;\mathrm {rem}}[V] , \nonumber \\ {\mathcal {I}}_{\pi ;b;U}[V]= & {} {\mathcal {I}}_{\pi ;b;U;\mathrm {sh}}[V] + {\mathcal {I}}_{\pi ;b;U;\mathrm {rem}}[V] . \end{aligned}$$
(4.59)

Following the same reasoning used to obtain [28,  Eq. (8.15)], we may use Propositions 4.2 and 4.3 to derive the bound

$$\begin{aligned} \left\| {\mathcal {I}}_{\pi ;a;U;\# }[V] \right\| _{\ell ^2_h}\le & {} C_1' \Big [ T_{\mathrm {safe}}(V) S_{\#;\mathrm {full}}(U) + T_{\infty ;\mathrm {opt}}(V) S_{\#;2;\mathrm {fix}}(U) \Big ] \nonumber \\\le & {} C_2' {\mathcal {E}}_{\#;U}(V) . \end{aligned}$$
(4.60)

Indeed, contributions of type \(T_{\infty ;\mathrm {opt}}(V) S_{\#;\mathrm {full}}(U)\) can be avoided by deviating from the preferred exponents judiciously.

In addition, following the arguments used to derive [28,  Eq. (8.12)], we may use Proposition 4.5 to obtain the bound

$$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {I}}_{\pi ;b;U;\# }[V] \right\| _{\ell ^2_h}\le & {} C_3' {\mathcal {E}}_{\#;U}(V). \end{array} \end{aligned}$$
(4.61)

Writing

$$\begin{aligned} \Delta _{b;i}= & {} \pi \Big [ f_{1;\mathrm {lin};U^{(2)} ;\mathrm {rem} }[V] - f_{1;\mathrm {lin};U^{(1)};\mathrm {rem} }[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U^{(2)} ) , \ldots , f_{k; \mathrm {apx};\mathrm {expl} }(U^{(2)} ) \Big ] , \nonumber \\ \Delta _{b;ii}= & {} \pi \Big [ f_{1;\mathrm {lin};U^{(1)} ;\mathrm {rem} }[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U^{(2)}) - f_{2;\mathrm {apx};\mathrm {expl}}(U^{(1)} ) , \ldots , f_{k; \mathrm {apx} ;\mathrm {expl} }(U^{(2)} ) \Big ] \nonumber \\&+ \ldots \nonumber \\&+ \pi \Big [ f_{1;\mathrm {lin};U^{(1)} ;\mathrm {rem} }[V] , f_{2;\mathrm {apx};\mathrm {expl}}(U^{(1)}), \ldots , f_{k;\mathrm {apx};\mathrm {expl}}(U^{(2)}) - f_{k;\mathrm {apx};\mathrm {expl}}(U^{(1)} ) \Big ] ,\nonumber \\ \end{aligned}$$
(4.62)

we easily see that

$$\begin{aligned} \Delta _{b;i} + \Delta _{b;ii} = {\mathcal {I}}_{\pi ;b;U^{(2)} ;\mathrm {rem} }[V] - {\mathcal {I}}_{\pi ;b;U^{(1)};\mathrm {rem} }[V]. \end{aligned}$$
(4.63)

Arguing as above, Proposition 4.6 yields

$$\begin{aligned} \begin{array}{lcl} \left\| \Delta _{b;i} \right\| _{\ell ^2_h}\le & {} C_1' {\mathcal {E}}_{\mathrm {prod}}(U^{(2)} - U^{(1)} , V), \end{array} \end{aligned}$$
(4.64)

while Propositions 4.2 and 4.4 imply

$$\begin{aligned} \left\| \Delta _{b;ii} \right\| _{\ell ^2_h}\le & {} C_2' \Big [ T_{\mathrm {safe}}(V) S_{\mathrm {diff};\mathrm {full}}(U^{(1)}, U^{(2)}) + T_{\infty ;\mathrm {opt}}(V) S_{\mathrm {diff};2;\mathrm {fix}}(U^{(1)}, U^{(2)}) \Big ] \nonumber \\\le & {} C_3' {\mathcal {E}}_{\mathrm {prod}}(U^{(2)} - U^{(1)},V). \end{aligned}$$
(4.65)

Finally, we write

$$\begin{aligned} \Delta _a = {\mathcal {I}}_{\pi ;a;U^{(2)};\mathrm {rem} }[V] - {\mathcal {I}}_{\pi ;a;U^{(1)};\mathrm {rem}}[V]. \end{aligned}$$
(4.66)

We note that \(\Delta _{a}\) consists of sums of expressions that arise from \(\Delta _{b;i}\) and \(\Delta _{b;ii}\) after replacing \(f_{1;\mathrm {lin};U^{(i)};\mathrm {rem}}\) by \(f_{1;\mathrm {lin};U^{(i)}}\) and each occurrence of \(f_{j;\mathrm {apx};\mathrm {expl}}\) by an element of the set

$$\begin{aligned} \{ f_{j;\mathrm {apx}}, f_{j;\mathrm {apx};\mathrm {expl}} , f_{j;\mathrm {apx};\mathrm {rem}} \}. \end{aligned}$$
(4.67)

We can hence again use Propositions 4.2, 4.4 and 4.6 to conclude that \(\left\| \Delta _a\right\| _{\ell ^2_h}\) can be bounded by terms that have already appeared above. The desired bounds now follow from the fact that \({\mathcal {G}}_{\mathrm {lin};U;I}[V] - {\mathcal {G}}_{\mathrm {lin};U;II}[V]\) can be written as a sum of expressions of the form \({\mathcal {I}}_{\pi ;a} + {\mathcal {I}}_{\pi ;b}\), together with their obvious permutations. \(\square \)

5 Estimates for \({\mathcal {G}}\) and \({\mathcal {G}}^+\)

In this section we exploit the component estimates from Sect. 4 to analyze the function \({\mathcal {G}}\) defined in (2.12), together with its first difference \({\mathcal {G}}^+\). In particular, recalling the operator \(L_U\) defined in (3.12), we introduce our final approximants

$$\begin{aligned} \begin{array}{lclclcl} {\mathcal {G}}_{\mathrm {apx}}(U) &{} = &{} c_* \partial ^0 U , &{} \qquad \qquad &{} {\mathcal {G}}_{\mathrm {lin};U}[V] &{} = &{} L_U[V], \\ {\mathcal {G}}^+_{\mathrm {apx}}(U) &{} = &{} \partial ^+ \big [ {\mathcal {G}}_{\mathrm {apx}}(U) \big ], &{} \qquad \qquad &{} {\mathcal {G}}^+_{\mathrm {lin};U}[V] &{} = &{} \partial ^+ \big [ {\mathcal {G}}_{\mathrm {lin};U}[V] \big ] \end{array} \end{aligned}$$
(5.1)

and write

$$\begin{aligned} {\mathcal {G}}_{\mathrm {nl};U}(V) = {\mathcal {G}}( U + V) - {\mathcal {G}}(U) - {\mathcal {G}}_{\mathrm {lin};U}(V) \end{aligned}$$
(5.2)

together with

$$\begin{aligned} {\mathcal {G}}^+_{\mathrm {nl};U}(V) = \partial ^+[ {\mathcal {G}}_{\mathrm {nl};U}(V)] = {\mathcal {G}}^+( U + V) - {\mathcal {G}}^+(U) - {\mathcal {G}}^+_{\mathrm {lin};U}(V) . \end{aligned}$$
(5.3)

Using the discrete calculus outlined in Sect. A, one may readily verify the identities

$$\begin{aligned} {\mathcal {G}}^+_{\mathrm {apx}}(U)= & {} c_* S^+[\partial ^{(2)} U] , \nonumber \\ {\mathcal {G}}^+_{\mathrm {lin};U}[V]= & {} c_* S^+[ \partial ^{(2)} V] + \partial ^+\big [ M_U[V] \big ] + \gamma _U^{-2} \partial ^0 U [\partial ^{(2)} U] M_U[V] \nonumber \\&+ S^+[\partial ^{(2)} U] T^+ \sum _{-;h} \gamma _U^{-2} \partial ^{(2)} U M_U[V] . \end{aligned}$$
(5.4)

Our main result quantifies the approximation errors in terms of the functions \({\mathcal {E}}_{\mathrm {sh};U}\), \({\mathcal {E}}_{\mathrm {rem};U}\) and \({\mathcal {E}}_{\mathrm {prod}}\) and their counterparts \(\overline{{\mathcal {E}}}_{\mathrm {sh};U}\), \(\overline{{\mathcal {E}}}_{\mathrm {rem};U}\) and \(\overline{{\mathcal {E}}}_{\mathrm {prod};U}\) defined in (3.26), (3.27), (3.28) and (3.34). For convenience, we also reference the quantities (4.16)–(4.17).

Proposition 5.1

Suppose that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). Then there exists \(K > 0\) so that the following properties hold.

  1. (i)

    For every \(h > 0\) and \(U \in \Omega _{h; \kappa }\) we have

    $$\begin{aligned} \left\| {\mathcal {G}}(U) - {\mathcal {G}}_{\mathrm {apx}}(U) \right\| _{\ell ^2_h}\le & {} K\big [ h + \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^2_h} + \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^\infty _h} \big ], \nonumber \\= & {} K S_{\mathrm {sh};\mathrm {full}} + K S_{\mathrm {rem};\mathrm {full}} \nonumber \\ \left\| {\mathcal {G}}^+(U) - {\mathcal {G}}^+_{\mathrm {apx}}(U) \right\| _{\ell ^2_h}\le & {} K h \big [ 1 + \left\| \partial ^+ \partial ^+ \partial ^+ U\right\| _{\ell ^2_h} + \left\| \partial ^+ \partial ^+ \partial ^+ U\right\| _{\ell ^\infty _h} \big ] \nonumber \\&\qquad + K \big [ \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^2_h} + \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^\infty _h} + \left\| \partial ^+ {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^2_h} \big ] \nonumber \\= & {} K {\overline{S}}_{\mathrm {sh};\mathrm {full}} + K {\overline{S}}_{\mathrm {rem};\mathrm {full}}. \end{aligned}$$
    (5.5)
  2. (ii)

    For any \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) for which \(U + V \in \Omega _{h;\kappa }\), we have the estimates

    $$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};U}(V) \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {prod}}( V, V) + K {\mathcal {E}}_{\mathrm {sh};U}(V) + K {\mathcal {E}}_{\mathrm {rem};U}(V), \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {nl};U}(V) \right\| _{\ell ^2_h}\le & {} K \overline{{\mathcal {E}}}_{\mathrm {prod};U}( V, V) + K h \overline{{\mathcal {E}}}_{\mathrm {sh};U}(V) + K \overline{{\mathcal {E}}}_{\mathrm {rem};U}(V) . \end{aligned}$$
    (5.6)
  3. (iii)

    Consider any \(h > 0\), \(U \in \Omega _{h;\kappa }\) and any pair \((V^{(1)}, V^{(2)} ) \in \ell ^2_h \times \ell ^2_h\) for which the inclusions \(U + V^{(1)} \in \Omega _{h;\kappa }\) and \(U + V^{(2)} \in \Omega _{h;\kappa }\) both hold. Then we have the Lipschitz estimate

    $$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};U}(V^{(2)}) - {\mathcal {G}}_{\mathrm {nl};U}(V^{(1)}) \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {prod}}(V^{(1)}, V^{(2)} - V^{(1)} ) + K {\mathcal {E}}_{\mathrm {prod}}(V^{(2)}, V^{(2)} - V^{(1)} ) \nonumber \\&+ K {\mathcal {E}}_{\mathrm {sh};U} ( V^{(2)} - V^{(1)} ) + K {\mathcal {E}}_{\mathrm {rem};U}(V^{(2)} - V^{(1)} ) . \end{aligned}$$
    (5.7)

5.1 Refinement Strategy

We recall the refined approximants \({\mathcal {G}}_{\mathrm {apx};II}(U)\) and \({\mathcal {G}}_{\mathrm {lin};U;II}[V]\) that we defined in Sect. 4.2. The main task in this section is to track the errors that accumulate as we reduce these expressions even further to our relatively simple approximants (5.1). In contrast to the abstract approach in Sect. 4, we achieve this in a direct fashion through several explicit computations. Indeed, in Sects. EF we obtain the following representations.

Proposition 5.2

(see Sects. EF) Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). There exists a constant \(K > 0\) together with sequences

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {apx};\mathrm {sh};b}(U) \in \ell ^2_h, \qquad {\mathcal {G}}_{\mathrm {apx};\mathrm {rem};b}(U) \in \ell ^2_h, \qquad {\mathcal {G}}^+_{\mathrm {apx};\mathrm {sh};b}(U) \in \ell ^2_h, \qquad {\mathcal {G}}^+_{\mathrm {apx};\mathrm {rem};b}(U) \in \ell ^2_h, \end{array}\nonumber \\ \end{aligned}$$
(5.8)

defined for every \(h > 0\) and \(U \in \Omega _{h;\kappa }\), so that the following properties hold true.

  1. (i)

    For every \(h > 0\) and \(U \in \Omega _{h;\kappa }\) we have the identities

    $$\begin{aligned} {\mathcal {G}}_{\mathrm {apx};II}(U)= & {} {\mathcal {G}}_{\mathrm {apx}}(U) + {\mathcal {G}}_{\mathrm {apx};\mathrm {sh};b}(U) + {\mathcal {G}}_{\mathrm {apx};\mathrm {rem};b}(U), \nonumber \\ {\mathcal {G}}^+_{\mathrm {apx};II}(U)= & {} {\mathcal {G}}^+_{\mathrm {apx}}(U) + {\mathcal {G}}^+_{\mathrm {apx};\mathrm {sh};b}(U) + {\mathcal {G}}^+_{\mathrm {apx};\mathrm {rem};b}(U). \end{aligned}$$
    (5.9)
  2. (ii)

    For every \(h > 0\) and \(U \in \Omega _{h;\kappa }\) we have the bounds

    $$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {apx};\mathrm {sh};b}(U) \right\| _{\ell ^2_h}\le & {} K S_{\mathrm {sh};\mathrm {full}}(U) \nonumber \\ \left\| {\mathcal {G}}_{\mathrm {apx};\mathrm {rem};b}(U) \right\| _{\ell ^2_h}\le & {} K S_{\mathrm {rem};\mathrm {full}}(U) , \end{aligned}$$
    (5.10)

    together with

    $$\begin{aligned} \left\| {\mathcal {G}}^+_{\mathrm {apx};\mathrm {sh};b}(U) \right\| _{\ell ^2_h}\le & {} K {\overline{S}}_{\mathrm {sh};\mathrm {full}}(U) , \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {apx};\mathrm {rem};b}(U) \right\| _{\ell ^2_h}\le & {} K {\overline{S}}_{\mathrm {rem};\mathrm {full}}(U) . \end{aligned}$$
    (5.11)

Proposition 5.3

(see Sects. EF) Assume that (Hg) is satisfied and fix \(0< \kappa < \frac{1}{12}\). There exists a constant \(K > 0\) together with linear maps

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};b} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ) , \qquad \qquad {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};b} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ) \end{array} \end{aligned}$$
(5.12)

and their counterparts

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};b} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ), \qquad \qquad {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};b} \in {\mathcal {L}}(\ell ^2_h, \ell ^2_h ), \end{array} \end{aligned}$$
(5.13)

defined for all \(h > 0\) and \(U \in \Omega _{h;\kappa }\), so that the following properties hold true.

  1. (i)

    For every \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) we have the identities

    $$\begin{aligned} {\mathcal {G}}_{\mathrm {lin};U;II}[V]= & {} {\mathcal {G}}_{\mathrm {lin};U}[V] + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};b}[V] + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};b}[V] , \nonumber \\ {\mathcal {G}}^+_{\mathrm {lin};U;II}[V]= & {} {\mathcal {G}}^+_{\mathrm {lin};U}[V] + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};b}[V] + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};b}[V] . \end{aligned}$$
    (5.14)
  2. (ii)

    For every \(h > 0\), \(U \in \Omega _{h;\kappa }\) and \(V \in \ell ^2_h\) we have the bounds

    $$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};b}[V] \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {sh};U}(V) , \nonumber \\ \left\| {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};b}[V] \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {rem};U}(V) , \end{aligned}$$
    (5.15)

    together with

    $$\begin{aligned} \left\| {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};b}[V] \right\| _{\ell ^2_h}\le & {} K \overline{{\mathcal {E}}}_{\mathrm {sh};U}(V) , \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};b}[V] \right\| _{\ell ^2_h}\le & {} K \overline{{\mathcal {E}}}_{\mathrm {rem};U}(V) . \end{aligned}$$
    (5.16)
  3. (iii)

    For every \(h > 0\), any pair \((U^{(1)}, U^{(2)}) \in \Omega _{h;\kappa }^2\) and any \(V \in \ell ^2_h\), we have the bound

    $$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {G}}_{\mathrm {lin};U^{(2)};\mathrm {rem};b}[V] - {\mathcal {G}}_{\mathrm {lin};U^{(1)};\mathrm {rem};b}[V] \right\| _{\ell ^2_h} &{} \le &{} K {\mathcal {E}}_{\mathrm {prod}}(U^{(2)} - U^{(1)} , V) . \\ \end{array} \end{aligned}$$
    (5.17)

Recalling the initial nonlinear residuals (3.50) together with the expressions (4.42), (4.51) and (4.52), we have hence obtained the decompositions

$$\begin{aligned} {\mathcal {G}}_{\mathrm {nl};U}(V)= & {} {\mathcal {G}}_{\mathrm {nl};U;I}(V) + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};a}[V] + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};b}[V] + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};a}[V] \nonumber \\&\quad + {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};b}[V] , \nonumber \\ {\mathcal {G}}^+_{\mathrm {nl};U}(V)= & {} {\mathcal {G}}^+_{\mathrm {nl};U;I}(V) + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};a}[V] + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {rem};b}[V] + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};a}[V] \nonumber \\&\quad + {\mathcal {G}}^+_{\mathrm {lin};U;\mathrm {sh};b}[V] . \end{aligned}$$
(5.18)

We now turn towards the Lipschitz bounds for \({\mathcal {G}}_{\mathrm {nl}}\).

Corollary 5.4

Assume that (Hg) is satisfied and pick \(0< \kappa < \frac{1}{12}\). There exists a constant \(K > 0\) so that the estimate

$$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {G}}_{\mathrm {lin};U^{(2)}}[V] - {\mathcal {G}}_{\mathrm {lin};U^{(1)} }[V] \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {prod}} \big (U^{(2)} - U^{(1)}, V \big ) \end{array} \end{aligned}$$
(5.19)

holds for all \(h > 0\), all \(V \in \ell ^2_h\) and all pairs \((U^{(1)}, U^{(2)} ) \in \Omega _{h;\kappa }^2\).

Proof

This is a direct restatement of [29,  Cor. 5.3]. \(\square \)

Lemma 5.5

Assume that (Hg) is satisfied and pick \(0< \kappa < \frac{1}{12}\). There exists a constant \(K > 0\) so that the estimate

$$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};U}(V^{(2)}) - {\mathcal {G}}_{\mathrm {nl};U}(V^{(1)}) \right\| _{\ell ^2_h}\le & {} K {\mathcal {E}}_{\mathrm {prod}} \big ( V^{(2)} - V^{(1)} , V^{(2)} - V^{(1)} \big ) \nonumber \\&+ K h \left\| V^{(2)} - V^{(1)} \right\| _{\ell ^{2;2}_h} \nonumber \\&+ K {\mathcal {E}}_{\mathrm {rem};U} \big ( V^{(2)} - V^{(1)} \big ) \nonumber \\&+ K {\mathcal {E}}_{\mathrm {prod}} \big ( V^{(1)} , V^{(2)} - V^{(1)} \big ) \end{aligned}$$
(5.20)

holds for all \(h > 0\), all \(U \in \Omega _{h;\kappa }\) and all pairs \((V^{(1)}, V^{(2)} ) \in \ell ^2_h \times \ell ^2_h\) for which the inclusions \(U + V^{(1)} \in \Omega _{h;\kappa }\) and \(U + V^{(2)} \in \Omega _{h;\kappa }\) both hold.

Proof

By definition, we have

$$\begin{aligned} \begin{array}{lcl} {\mathcal {G}}_{\mathrm {nl};U}(V)= & {} {\mathcal {G}}(U + V) - {\mathcal {G}}(U) - {\mathcal {G}}_{\mathrm {lin};U}[V] . \end{array} \end{aligned}$$
(5.21)

In particular, we get

$$\begin{aligned} {\mathcal {G}}_{\mathrm {nl};U}(V^{(2)}) - {\mathcal {G}}_{\mathrm {nl};U}(V^{(1)})= & {} {\mathcal {G}}(U + V^{(2)}) - {\mathcal {G}}_{\mathrm {lin}; U}[V^{(2)}] + {\mathcal {G}}_{\mathrm {lin}; U}[V^{(1)}] - {\mathcal {G}}(U + V^{(1)}) \nonumber \\= & {} {\mathcal {G}}\big (U + V^{(1)} + (V^{(2)} - V^{(1)})\big ) - {\mathcal {G}}(U + V^{(1)}) \nonumber \\&- {\mathcal {G}}_{\mathrm {lin}; U}[V^{(2)} - V^{(1)}] \nonumber \\= & {} {\mathcal {G}}_{\mathrm {lin}; U + V^{(1)} }[V^{(2)}-V^{(1)}] + {\mathcal {G}}_{\mathrm {nl}; U + V^{(1)}} (V^{(2)} - V^{(1)}) \nonumber \\&- {\mathcal {G}}_{\mathrm {lin}; U}[V^{(2)}-V^{(1)}] \nonumber \\= & {} {\mathcal {G}}_{\mathrm {nl}; U + V^{(1)}} (V^{(2)} - V^{(1)}) \nonumber \\&\qquad + \big [{\mathcal {G}}_{\mathrm {lin}; U + V^{(1)} } - {\mathcal {G}}_{\mathrm {lin}; U }\big ] [V^{(2)}-V^{(1)}] . \end{aligned}$$
(5.22)

For convenience, we write

$$\begin{aligned} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem}}[V]= & {} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};a}[V] +{\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem};b}[V], \nonumber \\ {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh}}[V]= & {} {\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};a}[V] +{\mathcal {G}}_{\mathrm {lin};U;\mathrm {sh};b}[V]. \end{aligned}$$
(5.23)

In view of (5.18), we find

$$\begin{aligned} {\mathcal {G}}_{\mathrm {nl};U}(V^{(2)}) - {\mathcal {G}}_{\mathrm {nl};U}(V^{(1)})= & {} {\mathcal {G}}_{\mathrm {nl}; U + V^{(1)}; I} (V^{(2)} - V^{(1)}) \nonumber \\&+ {\mathcal {G}}_{\mathrm {lin};U + V^{(1)};\mathrm {rem}} (V^{(2)} - V^{(1)}) + {\mathcal {G}}_{\mathrm {lin};U + V^{(1)};\mathrm {sh}} (V^{(2)} - V^{(1)}) \nonumber \\&+ \big [{\mathcal {G}}_{\mathrm {lin}; U + V^{(1)} } - {\mathcal {G}}_{\mathrm {lin}; U }\big ] [V^{(2)}-V^{(1)}] \nonumber \\= & {} {\mathcal {G}}_{\mathrm {nl}; U + V^{(1)}; I} (V^{(2)} - V^{(1)}) \nonumber \\&+ {\mathcal {G}}_{\mathrm {lin};U ;\mathrm {rem}} (V^{(2)} - V^{(1)}) + {\mathcal {G}}_{\mathrm {lin};U + V^{(1)};\mathrm {sh}} (V^{(2)} - V^{(1)})\nonumber \\&+\Big [{\mathcal {G}}_{\mathrm {lin};U + V^{(1)};\mathrm {rem}} - {\mathcal {G}}_{\mathrm {lin};U;\mathrm {rem}} \Big ] (V^{(2)} - V^{(1)}) \nonumber \\&+ \big [{\mathcal {G}}_{\mathrm {lin}; U + V^{(1)} } - {\mathcal {G}}_{\mathrm {lin}; U }\big ] [V^{(2)}-V^{(1)}] . \end{aligned}$$
(5.24)

The desired bound now follows from (3.51), Lemma’s 4.7-4.8, Propositions 5.2-5.3 and Corollary 5.4. \(\square \)

Proof

In view of the expression (5.18), the statements follow from (3.49), (3.51), Lemma’s 4.74.8, Propositions 5.25.3 and Lemma 5.5. \(\square \)

6 Travelling Waves

Formally substituting the travelling wave Ansatz (2.20) into the reduced system (2.13) leads to the nonlocal differential equation

$$\begin{aligned} c \Psi ' = {\mathcal {G}}(\Psi ). \end{aligned}$$
(6.1)

In this section we set out to construct solutions to this equation for small \(h > 0\) that can be written as

$$\begin{aligned} \Psi = \Psi _* + v, \qquad c = c_* + {\tilde{c}} \end{aligned}$$
(6.2)

for pairs \(({\tilde{c}}, v)\) that tend to zero as \(h \downarrow 0\). Care must be taken to ensure that the expression \({\mathcal {G}}(\Psi )\) is well-defined, but based on our preparations we are able to provide a relatively streamlined fixed-point argument here, which allows us to prove the results stated in Sect. 2.

In order to control the size of the perturbation \(({\tilde{c}}, v) \in {\mathbb {R}}\times H^1\), we introduce the norms

$$\begin{aligned} \left\| ({\tilde{c}},v)\right\| _{{\mathcal {Z}}_h } = \left| {\tilde{c}}\right| + \left\| v\right\| _{H^1} + \left\| \partial ^+_h \partial ^+_h v\right\| _{L^2} \end{aligned}$$
(6.3)

for \(h > 0\) and write \({\mathcal {Z}}_h\) for the set \({\mathbb {R}}\times H^1\) equipped with this new norm. Observe that for fixed h this norm is equivalent to the usual one on \({\mathbb {R}}\times H^1\).

Recalling the discussion at the start of Sect. 3, we pick \(0< \kappa < \frac{1}{12}\) and \(\epsilon _0 > 0\) in such a way that the inclusion

$$\begin{aligned} \mathrm {ev}_{\vartheta } [\Psi _* + v ] \in \Omega _{h;\kappa } \end{aligned}$$
(6.4)

holds for all \(0< h < 1\), all \(\vartheta \in [0, h]\) and all \(v \in H^1\) that satisfy (3.8). In order to accommodate this and control the third-differences of v, we pick two parameters \(\delta > 0\) and \(\delta _v^+ > 0\) and introduce the set

$$\begin{aligned} {\mathcal {Z}}_{h; \delta , \delta _v^+}= & {} \big \{ ({\tilde{c}},v ) \in {\mathcal {Z}}_h : \left\| ({\tilde{c}},v)\right\| _{{\mathcal {Z}}_h} \le \min \{ \delta , \epsilon _0 \} \nonumber \\&\hbox { and } \left\| (0, \partial ^+_h v ) \right\| _{{\mathcal {Z}}_h} \le \min \{ \delta _v^+ , h^{1/2} \epsilon _0 \} \big \}. \end{aligned}$$
(6.5)

Since \(\partial ^+_h\) is bounded on \(H^1\) and \(L^2\) for each fixed h, we note that this is a closed subset of \({\mathcal {Z}}_h\).

Substituting (6.2) into (6.1), we obtain

$$\begin{aligned} c_* \Psi _*' + {\tilde{c}} \Psi _*' + {\tilde{c}} v' + c_* v'= & {} {\mathcal {G}}(\Psi _* + v) \nonumber \\= & {} {\mathcal {G}}(\Psi _*) + {\mathcal {G}}_{\mathrm {lin};\Psi _*}[v] + {\mathcal {G}}_{\mathrm {nl};\Psi _*}(v), \end{aligned}$$
(6.6)

which should be interpreted in a sense similar to that of (3.18).

Upon introducing the nonlinearity

$$\begin{aligned} \begin{array}{lcl} {\mathcal {H}}_{h}({\tilde{c}}, v)= & {} {\tilde{c}} v' - {\mathcal {G}}_{\mathrm {nl};\Psi _*}(v) \end{array} \end{aligned}$$
(6.7)

and inspecting the definitions (3.16) and (5.1), we can rewrite (6.6) as

$$\begin{aligned} {\mathcal {L}}_{ h } [v]= & {} {\tilde{c}} \Psi _*' + {\mathcal {H}}_{h} ({\tilde{c}}, v) + c_* \Psi _*' - {\mathcal {G}}(\Psi _*) . \end{aligned}$$
(6.8)

Recalling the two solution operators (3.19), we now introduce the map \({\mathcal {W}}_h: {\mathcal {Z}}_{h;\delta ,\delta _v^+} \rightarrow {\mathcal {Z}}_h\) that acts as

$$\begin{aligned} \begin{array}{lcl} {\mathcal {W}}_h({\tilde{c}}, v)= & {} [\beta ^*_h , {\mathcal {V}}^*_h] \Big [ {\mathcal {H}}_{h} ({\tilde{c}}, v) + c_* \Psi _*' - {\mathcal {G}}(\Psi _*) \Big ] , \end{array} \end{aligned}$$
(6.9)

which allows us to recast (6.8) as the fixed point problem

$$\begin{aligned} ({\tilde{c}},v) = {\mathcal {W}}_h\big ({\tilde{c}}, v\big ) . \end{aligned}$$
(6.10)

In order to show that \({\mathcal {W}}_h\) is a contraction mapping on \({\mathcal {Z}}_{h;\delta ,\delta _v^+}\), we study the two expressions \({\mathcal {H}}_h\) and \(c_* \Psi _*' - {\mathcal {G}}(\Psi _*)\) separately in our first results. The sampling bounds from Sect. A play a key role here, as they enable us to extract \(L^2\)-based bounds on \({\mathcal {G}}_{\mathrm {nl};\Psi _*}\) from the sequence estimates obtained in Sect. 5.

Notice that the control (6.15) on \(\left\| v\right\| _{\ell ^{2;2}_h}\) would not have been possible using only bounds on (6.3), since \(L^2\)-norms cannot directly be turned into \(\ell ^2_h\)-norms. This would prevent us from bounding the terms that are quadratic in these second differences. In fact, this is the reason that we needed to obtain such detailed bounds on \({\mathcal {G}}^+\) in this series of papers. Indeed, the additional third-differences only appear in a linear fashion, which does allow us to easily pass between sequences and functions; see (A.18).

Lemma 6.1

Suppose that (Hg) and (H\(\Phi _*\)) are satisfied. There exists \(K >0\) so that for any pair \((\delta , \delta _v^+) \in (0, 1)^2\) and any \(0< h < 1\) the estimates

$$\begin{aligned} \left\| {\mathcal {H}}_h({\tilde{c}}, v) \right\| _{L^2}\le & {} K \big [ h \delta + \delta ^2 + \delta \delta _v^+ \big ] , \nonumber \\ \left\| \partial ^+ {\mathcal {H}}_h({\tilde{c}}, v) \right\| _{L^2}\le & {} K \big [ [\delta + \delta _v^+ ]^2 + h^{-1/2} \delta [\delta + \delta ^+_v] + h [\delta _v + \delta _v^+ ] \big ] \end{aligned}$$
(6.11)

hold for each \(({\tilde{c}},v) \in {\mathcal {Z}}_{h;\delta ,\delta _v^+}\), while the estimate

$$\begin{aligned} \left\| {\mathcal {H}}_h({\tilde{c}}^{(2)}, v^{(2)}) - {\mathcal {H}}_h({\tilde{c}}^{(1)}, v^{(1)}) \right\| _{L^2}\le & {} K \big [ h^{-1/2} [ \delta + \delta _v^+ ] + h \big ] \left\| ({\tilde{c}}^{(2)} - {\tilde{c}}^{(1)}, v^{(2)} - v^{(1)} ) \right\| _{{\mathcal {Z}}_h}\nonumber \\ \end{aligned}$$
(6.12)

holds for each set of pairs \(({\tilde{c}}^{(1)},v^{(1)}) \in {\mathcal {Z}}_{h;\delta ,\delta _v^+}\) and \(({\tilde{c}}^{(2)},v^{(2)}) \in {\mathcal {Z}}_{h;\delta ,\delta _v^+}\).

Proof

The first term in \({\mathcal {H}}_h\) can be handled by the elementary estimates

$$\begin{aligned} \left\| {\tilde{c}} v'\right\| _{L^2}\le & {} \delta ^2,\nonumber \\ \left\| {\tilde{c}} \partial ^+ v'\right\| _{L^2}\le & {} \delta ,\nonumber \\ \left\| \partial ^+ v\right\| _{H^1}\le & {} \delta \delta _v^+ , \end{aligned}$$
(6.13)

together with

$$\begin{aligned} \left\| {\tilde{c}}^{(2)} [v^{(2)}]' - {\tilde{c}}^{(1)}[v^{(1)}]' \right\| _{L^2}\le & {} \left| {\tilde{c}}^{(2)} - {\tilde{c}}^{(1)} \right| \left\| v^{(2)}\right\| _{H^1} + \left| {\tilde{c}}^{(1)} \right| \left\| v^{(1)} - v^{(2)} \right\| _{H^1} \nonumber \\\le & {} \delta \left\| \big ( {\tilde{c}}^{(2)} - {\tilde{c}}^{(1)} , v^{(2)} - v^{(1)} \big ) \right\| _{{\mathcal {Z}}_h}. \end{aligned}$$
(6.14)

Using Corollary A.1 we see that

$$\begin{aligned} \left\| v\right\| _{\ell ^{2;2}_h} + \left\| v\right\| _{\ell ^{\infty ;1}_h} \le C_1' [\delta + \delta _v^+ ] \end{aligned}$$
(6.15)

for all \(({\tilde{c}},v) \in {\mathcal {Z}}_{h;\delta ,\delta ^+}\). For any \(\vartheta \in {\mathbb {R}}\), we may hence exploit Propositions C.3 and 5.1 to obtain the estimate

$$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};\Psi _*}( \mathrm {ev}_{\vartheta } v ) \right\| _{\ell ^2_h}\le & {} C_2' \big [ \delta + \delta _v^+ + h \big ] \left\| \mathrm {ev}_{\vartheta } v\right\| _{\ell ^{2;2}_h}, \end{aligned}$$
(6.16)

together with

$$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};\Psi _*}( \mathrm {ev}_{\vartheta } v^{(2)} ) - {\mathcal {G}}_{\mathrm {nl};\Psi _*}(\mathrm {ev}_{\vartheta } v^{(1)} ) \right\| _{\ell ^2_h}\le & {} C_2' \big [ \delta + \delta _v^+ + h \big ] \left\| \mathrm {ev}_{\vartheta } v^{(1)} - \mathrm {ev}_{\vartheta } v^{(2)}\right\| _{\ell ^{2;2}_h} \nonumber \\&+ C_2'\big [ \delta + \delta _v^+ ] \left\| \mathrm {ev}_{\vartheta } v^{(1)} - \mathrm {ev}_{\vartheta } v^{(2)}\right\| _{\ell ^{\infty ;1}_h} .\nonumber \\ \end{aligned}$$
(6.17)

A second application of Corollary A.1 yields the bound

$$\begin{aligned} \left\| v\right\| _{\ell ^{2;2}_h} + \left\| v\right\| _{\ell ^{\infty ;2}_h } \le C_3' h^{-1/2} [\delta + \delta ^+_v ] . \end{aligned}$$
(6.18)

For any \(\vartheta \in {\mathbb {R}}\), we may hence use Propositions C.3 and 5.1 to find

$$\begin{aligned} \left\| {\mathcal {G}}^+_{\mathrm {nl};\Psi _*}( \mathrm {ev}_{\vartheta } v ) \right\| _{\ell ^2_h}\le & {} C_4' \big [ \delta + \delta _v^+ + h \big ] \left\| \mathrm {ev}_{\vartheta } v\right\| _{\ell ^{2;3}_h} + C_4' h^{-1/2} [\delta + \delta _v^+ ] \left\| \mathrm {ev}_{\vartheta } v\right\| _{\ell ^{2;2}_h} .\nonumber \\ \end{aligned}$$
(6.19)

We now apply Lemma A.2 to obtain

$$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};\Psi _*}( v ) \right\| _{L^2}\le & {} C_2' \big [ \delta + \delta _v^+ + h \big ] \big [ \left\| v\right\| _{H^1} + \left\| \partial ^+ \partial ^+ v\right\| _{L^2} \big ] \nonumber \\\le & {} C_2' \big [ \delta + \delta _v^+ + h \big ] \delta , \nonumber \\ \left\| {\mathcal {G}}^+_{\mathrm {nl};\Psi _*}( v ) \right\| _{L^2}\le & {} C_4' \big [ \delta + \delta _v^+ + h \big ] \big [ \left\| v\right\| _{H^1} + \left\| \partial ^+ \partial ^+ v\right\| _{L^2} + \left\| \partial ^+ v\right\| _{H^1} + \left\| \partial ^+ \partial ^+ \partial ^+ v\right\| _{L^2} \big ] \nonumber \\&+ C_4' h^{-1/2} [\delta + \delta _v^+ ] \big [ \left\| v\right\| _{H^1} + \left\| \partial ^+ \partial ^+ v\right\| _{L^2} \big ] \nonumber \\\le & {} C_4' \big [ \delta + \delta _v^+ + h \big ] \big [ \delta + \delta _v^+ \big ] + C_4' h^{-1/2} [\delta + \delta _v^+ ] \delta . \end{aligned}$$
(6.20)

Using (A.6) we note that

$$\begin{aligned} \left\| v^{(2)} - v^{(1)}\right\| _{\ell ^{\infty ;1}_h } \le 2 h^{-1/2} \left\| v^{(2)} - v^{(1)}\right\| _{H^1} . \end{aligned}$$
(6.21)

Applying Lemma A.2 once more, we obtain

$$\begin{aligned} \left\| {\mathcal {G}}_{\mathrm {nl};\Psi _*}( v^{(2)} ) - {\mathcal {G}}_{\mathrm {nl};\Psi _*}( v^{(1)} ) \right\| _{L^2}\le & {} C_2' \big [ \delta + \delta _v^+ + h \big ] \Big [ \left\| v^{(1)} - v^{(2)}\right\| _{H^1} \nonumber \\&+ \left\| \partial ^+ \partial ^+ v^{(1)} - \partial ^+ \partial ^+ v^{(2)}\right\| _{L^2} \Big ] \nonumber \\&+ 2 C_2'\big [ \delta + \delta _v^+ ] h^{-1/2} \left\| v^{(1)} - v^{(2)}\right\| _{H^1}. \end{aligned}$$
(6.22)

The desired bounds follow readily from these estimates. \(\square \)

Lemma 6.2

Suppose that (Hg) and (H\(\Phi _*\)) are satisfied. There exists \(K > 0\) so that for each \(0< h < 1\) we have the bounds

$$\begin{aligned} \left\| c_* \Psi _*' - {\mathcal {G}}(\Psi _*) \right\| _{L^2}\le & {} K h , \nonumber \\ \left\| \partial ^+\big [ c_* \Psi _*' - {\mathcal {G}}(\Psi _*) \big ] \right\| _{L^2}\le & {} K h . \end{aligned}$$
(6.23)

Proof

Applying Lemma A.2 together with Propositions C.3 and 5.1, we find

$$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {G}}_{\mathrm {apx}}(\Psi _*) - {\mathcal {G}}(\Psi _*) \right\| _{L^2} + \left\| {\mathcal {G}}^+_{\mathrm {apx}}(\Psi _*) - {\mathcal {G}}^+(\Psi _*) \right\| _{L^2}\le & {} C_1' h . \end{array} \end{aligned}$$
(6.24)

We now compute

$$\begin{aligned} c_* \Psi _*' - {\mathcal {G}}(\Psi _*)= & {} c_* \Psi _*' - {\mathcal {G}}_{\mathrm {apx}}(\Psi _*) + {\mathcal {G}}_{\mathrm {apx}}(\Psi _*) - {\mathcal {G}}(\Psi _*) \nonumber \\= & {} c_* \Psi _*' - c_* \partial ^0 \Psi _* + {\mathcal {G}}_{\mathrm {apx}}(\Psi _*) - {\mathcal {G}}(\Psi _*) , \end{aligned}$$
(6.25)

together with

$$\begin{aligned} \partial ^+ [c_* \Psi _*' - {\mathcal {G}}(\Psi _*) \big ]= & {} \partial ^+ \big [ c_* \Psi _*' - {\mathcal {G}}_{\mathrm {apx}}(\Psi _*) \big ] + \partial ^+ \big [ {\mathcal {G}}_{\mathrm {apx}}(\Psi _*) - {\mathcal {G}}(\Psi _*) \big ] \nonumber \\= & {} c_* \big [ \partial ^+ \Psi _*]' - c_* \partial ^0 [\partial ^+ \Psi _*] + {\mathcal {G}}^+_{\mathrm {apx}}(\Psi _*) - {\mathcal {G}}^+(\Psi _*) . \end{aligned}$$
(6.26)

Applying (A.8) we see that

$$\begin{aligned} \begin{array}{lclcl} \left\| c_* \Psi _*' - {\mathcal {G}}(\Psi _*)\right\| _{L^2}\le & {} C_2' h \left\| \Psi _*''\right\| _{L^2} + C_1' h\le & {} C_3' h, \end{array} \end{aligned}$$
(6.27)

together with

$$\begin{aligned} \begin{array}{lclcl} \left\| \partial ^+ [c_* \Psi _*' - {\mathcal {G}}(\Psi _*)] \right\| _{L^2}\le & {} C_2' h \left\| \partial ^+ \Psi _*''\right\| _{L^2} + C_1' h\le & {} C_3'h , \end{array} \end{aligned}$$
(6.28)

as desired. \(\square \)

Utilizing the linear theory from [29] that we outlined in Sect. 3.1, we are now in a position to study the full nonlinear term \({\mathcal {W}}_h\). Our main result subsequently follows in a relatively standard fashion from the uniqueness properties of the contraction mapping theorem.

Lemma 6.3

Suppose that (Hg) and (H\(\Phi _*\)) are satisfied. Then for each sufficiently small \(h > 0\), the fixed point problem (6.10) posed on the set \({\mathcal {Z}}_{h; h^{3/4}, h^{3/4} }\) has a unique solution.

Proof

Using the estimates (3.22) together with the a-priori bounds \((h, \delta , \delta _v^+) \in (0, 1)^3\), we obtain the inequalities

$$\begin{aligned} \left\| {\mathcal {W}}_h({\tilde{c}}, v) \right\| _{{\mathcal {Z}}_h}\le & {} C_1' \big [ \left\| {\mathcal {H}}_{h}({\tilde{c}} , v) \right\| _{L^2} + \left\| c_* \Psi _*'- {\mathcal {G}}(\Psi _*)\right\| _{L^2} \big ] \nonumber \\\le & {} C_2' \big [ \delta ^2 + \delta \delta _v^+ + h \big ] , \nonumber \\ \left\| [0, \partial ^+ ] {\mathcal {W}}_h({\tilde{c}}, v) \right\| _{{\mathcal {Z}}_h}\le & {} C_1' \big [ \left\| {\mathcal {H}}_{h}({\tilde{c}} , v) \right\| _{L^2} +\left\| \partial ^+ {\mathcal {H}}_{h}({\tilde{c}} , v)\right\| _{L^2} \big ] \nonumber \\&+ C_1' \Big [ \left\| c_* \Psi _*'- {\mathcal {G}}(\Psi _*)\right\| _{L^2} + \left\| \partial ^+ \big [ c_* \Psi _*'- {\mathcal {G}}(\Psi _*) \big ]\right\| _{L^2} \Big ] \nonumber \\\le & {} C_2' \Big [ h^{-1/2} \delta [\delta + \delta _v^+] + (\delta + \delta _v^+)^2 + h \Big ] , \end{aligned}$$
(6.29)

together with

$$\begin{aligned} \left\| {\mathcal {W}}_h\big ( {\tilde{c}}^{(2)} , v^{(2)} \big ) - {\mathcal {W}}_h\big ( {\tilde{c}}^{(1)} , v^{(1)} \big ) \right\| _{{\mathcal {Z}}_h}\le & {} C_1' \left\| {\mathcal {H}}_{h}({\tilde{c}}^{(2)} , v^{(2)}) - {\mathcal {H}}_{h}({\tilde{c}}^{(1)} , v^{(1)}) \right\| _{L^2} \nonumber \\\le & {} C_2' \big [ h^{-1/2} [ \delta + \delta _v^+ ] + h \big ] \nonumber \\&\left\| \big ( {\tilde{c}}^{(2)} - {\tilde{c}}^{(1)} , v^{(2)} - v^{(1)} \big ) \right\| _{{\mathcal {Z}}_h} . \end{aligned}$$
(6.30)

Picking

$$\begin{aligned} \delta = \delta _v^+ = h^{3/4}, \end{aligned}$$
(6.31)

we see that \(\delta = \delta _v^+ \le h^{1/2} \epsilon _0\) for all sufficiently small \(h > 0\). In addition, we find

$$\begin{aligned} \left\| {\mathcal {W}}_h({\tilde{c}}, v) \right\| _{{\mathcal {Z}}_h}\le & {} C_2' \big [ 2 h^{3/4} + h^{1/4} \big ] \delta , \nonumber \\ \left\| [0, \partial ^+ ] {\mathcal {W}}_h({\tilde{c}}, v) \right\| _{{\mathcal {Z}}_h}\le & {} C_2' \Big [ 2 h^{1/4} + 4 h^{3/4} + h^{1/4} \Big ] \delta , \end{aligned}$$
(6.32)

together with

$$\begin{aligned} \begin{array}{lcl} \left\| {\mathcal {W}}_h \big ({\tilde{c}}^{(2)} , v^{(2)} \big ) - {\mathcal {W}}_h\big ( {\tilde{c}}^{(1)} , v^{(1)} \big ) \right\| _{{\mathcal {Z}}_h}\le & {} C_2'[2 h^{1/4} + h ] \left\| \big ( {\tilde{c}}^{(2)} - {\tilde{c}}^{(1)} , v^{(2)} - v^{(1)} \big ) \right\| _{{\mathcal {Z}}_h} . \end{array}\nonumber \\ \end{aligned}$$
(6.33)

The result hence follows from the contraction mapping theorem. \(\square \)

Proof of Theorem 2.1

We write \(({\tilde{c}}_h, v_h)\) for the unique solution to the fixed point problem (6.10) that is provided by Lemma 6.3. This allows us to define

$$\begin{aligned} \Psi _h = \Psi _* + v_h, \qquad \qquad c_h = c_ * + {\tilde{c}}_h . \end{aligned}$$
(6.34)

For fixed \(h > 0\), we claim that the map

$$\begin{aligned} \vartheta \mapsto \mathrm {ev}_{\vartheta } [\Psi _* + v_h ] - \mathrm {ev}_0 \Psi _* \in \ell ^2_h \end{aligned}$$
(6.35)

is continous. Indeed, this follows from the smoothness of \(\Psi _*\) together with (A.6) and the fact that the translation operator is continuous on \(H^1\). Since the map

$$\begin{aligned} V \mapsto {\mathcal {G}}(\Psi _* + V) \in \ell ^2_h \end{aligned}$$
(6.36)

is continuous on a subset of \(\ell ^2_h\) that contains \(\mathrm {ev}_{\vartheta } v_h\) for all \(\vartheta \in [0, h]\), we conclude that

$$\begin{aligned} \vartheta \mapsto {\mathcal {G}}\big ( \mathrm {ev}_{\vartheta } \Psi _h \big ) \in \ell ^2_h \end{aligned}$$
(6.37)

is continuous. The travelling wave equation (6.1) now implies the inclusion (2.21).

In a similar fashion, the inclusion (2.25) follows from (A.6) and the continuity of the translation operator on \(H^1\). The remaining statements are a direct consequence of Lemma 6.3. \(\square \)

We now turn to the proof of Corollary 2.2, which asserts the existence of a waveprofile \(\Phi _h\) in the original physical coordinates. The key tool for our purpose here is [28,  Prop. 4.2], which states that the gridpoints associated to a solution U of (2.13) satisfy

$$\begin{aligned} {\dot{x}}(t) = {\mathcal {M}}\big ( U(t) \big ). \end{aligned}$$
(6.38)

Here the sequence \({\mathcal {M}}\) can be written as

$$\begin{aligned} {\mathcal {M}}(U)= & {} - {\mathcal {Z}}^-(U) T^-[ {\mathcal {X}}_A(U) ]{\mathcal {Y}}_2(U) + {\mathcal {Z}}^-(U) \sum _{-;h} {\mathcal {Y}}_2(U) T^-\big [ {\mathcal {X}}_B(U) \big ] {\mathcal {D}}^{\diamond _-;+}(U) \nonumber \\&+{\mathcal {Z}}^-(U) \sum _{-;h} {\mathcal {Y}}_2(U) T^- \big [ {\mathcal {X}}_{C} (U) {\mathcal {D}}^{\diamond _0;+}(U) + {\mathcal {X}}_{D} (U) {\mathcal {D}}^{\diamond _0;+}(U) \big ]; \end{aligned}$$
(6.39)

see [28,  Eq. (6.31) and (6.33)] where this function was referred to as \({\mathcal {Y}}\). Notice the strong resemblance with the structure of (3.42). Indeed, we see that

$$\begin{aligned} {\mathcal {Y}}_1(U){\mathcal {M}}(U) = {\mathcal {Z}}^-(U) [{\mathcal {G}}(U) - {\mathcal {Y}}_2(U)] \end{aligned}$$
(6.40)

see also [28,  Eq. (6.9)] for comparison. In view of the identities

$$\begin{aligned}&{\mathcal {Z}}^-_{\mathrm {apx}}(U) = \gamma _U, \qquad {\mathcal {Y}}_{1;\mathrm {apx}}(U) = \partial ^0 U, \nonumber \\&{\mathcal {G}}_{\mathrm {apx}}(U) = c_* \partial ^0 U,\; {\mathcal {Y}}_{2;\mathrm {apx};\mathrm {expl}}(U) = c_* \gamma _U^{-1} \partial ^0 U \end{aligned}$$
(6.41)

from [28,  Eq. (7.29)], (D.1), (5.1) and (D.11), it makes sense to formally factor out \(\partial ^0 U\) and introduce the approximant

$$\begin{aligned} {\mathcal {M}}_{\mathrm {apx}}(U) = \gamma _U c_* (1 - \gamma _U^{-1} ) = c_* ( \gamma _U - 1). \end{aligned}$$
(6.42)

This allows us to extract a crucial lower bound for the speed of the gridpoints.

Proof of Corollary 2.2

Upon defining

$$\begin{aligned} \Xi _h = - \sum _{-;h} \frac{(\partial ^+\Psi _h)^2}{\sqrt{ 1 - (\partial ^+ \Psi _h)^2} + 1 }, \end{aligned}$$
(6.43)

the identity (2.27) implies that (i) is satisfied. Using [28,  Prop. 4.2] we see that

$$\begin{aligned} {\dot{x}}_{jh}(t) = c_h \Xi _h'(jh + c_h t) = \big [{\mathcal {M}}\big (\Psi _h(\cdot + c_h t) \big ) \big ]_{jh}. \end{aligned}$$
(6.44)

Notice that the terms appearing in (6.39) also all appear in (3.42) after making the replacement \({\mathcal {Z}}^- \mapsto {\mathcal {Y}}_1\). The reduction \({\mathcal {Z}}^- \mapsto {\mathcal {Z}}^-_{\mathrm {apx}}\) leads to error terms that are covered by the theory in [28]. Since \({\mathcal {Z}}^-_{\mathrm {apx}}\) does not need to be reduced further, we can follow all the computations in the present paper to obtain the error bound

$$\begin{aligned} \left\| {\mathcal {M}}(U) - {\mathcal {M}}_{\mathrm {apx}}(U)\right\| _{\ell ^\infty _h} \le C_1' \big [ h + \left\| {\mathcal {E}}_{\mathrm {tw}}(U)\right\| _{\ell ^2_h} + \left\| {\mathcal {E}}_{\mathrm {tw}}(U) \right\| _{\ell ^\infty _h} \big ], \end{aligned}$$
(6.45)

which is the natural analogue of (5.5). Substituting \(U = \Psi _h\) and applying the Lipschitz bounds (C.6), we find

$$\begin{aligned} \left\| {\mathcal {E}}_{\mathrm {tw}}(\Psi _h) - {\mathcal {E}}_{\mathrm {tw}}(\Psi _*) \right\| _{\ell ^2_h}\le & {} C_2' \left\| \Psi _h - \Psi _*\right\| _{\ell ^{2;2}_h} \nonumber \\\le & {} C_2' \big [ \left\| \Psi _h - \Psi _*\right\| _{H^1} + \left\| \partial ^+ [\Psi _h - \Psi _* ] \right\| _{H^1} \big ] \nonumber \\\le & {} C_2' h^{3/4}. \end{aligned}$$
(6.46)

Using Proposition C.3, we obtain

$$\begin{aligned} \left\| {\mathcal {E}}_{\mathrm {tw}}(\Psi _h)\right\| _{\ell ^2_h} \le C_3' h^{3/4} \end{aligned}$$
(6.47)

and hence

$$\begin{aligned} \left\| {\mathcal {E}}_{\mathrm {tw}}(\Psi _h)\right\| _{\ell ^\infty _h} \le C_3' h^{1/4}. \end{aligned}$$
(6.48)

In a similar fashion, we may exploit (B.4) to conclude

$$\begin{aligned} \left\| \gamma _{\Psi _h} - \gamma _{\Psi _*} \right\| _{\ell ^2_h} \le C_2' h^{3/4} \end{aligned}$$
(6.49)

and hence

$$\begin{aligned} \left\| \gamma _{\Psi _h} - \gamma _{\Psi _*} \right\| _{\ell ^\infty _h} \le C_2' h^{1/4}. \end{aligned}$$
(6.50)

Together, these observations yield the pointwise bound

$$\begin{aligned} \left| {\mathcal {M}}(\Psi _h) - c_* (\gamma _{\Psi _*} - 1)\right| \le C_4' h^{1/4}. \end{aligned}$$
(6.51)

Assuming for clarity that \(c_* > 0\), this implies the pointwise inequality

$$\begin{aligned} {\mathcal {M}}(\Psi _h) > c_* (\gamma _{\Psi _*} - 1) - C_4' h^{1/4}. \end{aligned}$$
(6.52)

Since \(\left| c_h - c_*\right| \le h^{3/4}\), we find

$$\begin{aligned} c_h \Big [ \Xi _h' + 1 \Big ]> & {} c_* (\gamma _{\Psi _*} - 1) + c_h - C_4' h^{1/4} \nonumber \\> & {} c_* \gamma _{\Psi _*} - C_5' h^{1/4}. \end{aligned}$$
(6.53)

Since \(\gamma _{\Psi _*}\) is strictly bounded away from zero, uniformly in h, we conclude that

$$\begin{aligned} \Xi _h'(\tau ) > -1 \end{aligned}$$
(6.54)

for all sufficiently small \(h > 0\) and all \(\tau \in {\mathbb {R}}\). This shows that the coordinate transformation (2.32) is invertible, as desired. \(\square \)