Long Time Behaviour of the Discrete Volume Preserving Mean Curvature Flow in the Flat Torus

Let h^*>0, \delta^*>0 and (\tau_n)_{n \in\mathbb{N}} be given by Proposition 5.2. Fix h<h^* and set E_n:=E_h^n. We split the proof in three steps.
Step 1. (Exponential convergence of dissipations) Testing the minimality of E_n with E_{n-1} we obtain

Recalling that P(E_n)\to P(E) and summing the previous inequality from n+1 to +\infty we get

We will now construct a suitable competitor to estimate the dissipation at the step n-1 with the difference of perimeters. Since, by Proposition 5.2, we have

the translated sets of the flow, for n large enough, can be written as normal deformations of the set E, that is there exists g_n : \partial E\to \mathbb{R} such that

where E_{g_n} was defined in (2.1). The convergence (?) then reads as g_n\to 0 in C^k as n\to \infty. Let \sigma_n be the translations introduced by Lemma 2.10 with E_n +\tau_n instead of F. From the convergence in C^k of E_n+\tau_n to E, we deduce that \sigma_n \to 0 as n\to\infty. Therefore, setting

we have that F_n\to E in C^k and F_n=E_{f_n} with f_n:\partial E\to\mathbb{R} satisfying

for p>N-1. Consider now the competitor

From the minimality of E_n we easily deduce

where we used the translational invariance of the dissipations. From Lemma 5.1 we obtain that the sequence E_{n-2}+\tau_{n-1}+\sigma_{n-1} converges in C^k to the same limit of E_{n-1}+\tau_{n-1}+\sigma_{n-1}, that is to E. In particular, for n large enough we can write E_{n-2}+\tau_{n-1}+\sigma_{n-1}=E_\psi for a suitable function \psi:\partial E\to \mathbb{R} (depending on n) and with \|\psi\|_{C^1(\partial E)} small. From Lemma 5.4 we can then estimate the right hand side of (?) with

From the previous inequality and (?) we obtain

Now, (Proof) and (?) yield

We then apply Lemma 5.5 below (with l=2) to conclude

Step 2. (Exponential convergence of barycenters) Set

From (?) and Lemma 5.1 both the sequences (E_n+\tau_n)_{n\in\mathbb{N}} and (E_{n-1}+\tau_n)_{n\in\mathbb{N}} converge in C^k to E. Therefore, for n large enough, there exist some functions f_{1,n},\, f_{2,n}\in C^k(\partial E) such that

and \|f_{i,n}\|_{C^k(\partial E)}\to 0 as n\to \infty for i=1,2. From (5.3) and (?) we can estimate for n sufficiently large

In turn, the above estimate implies that (\text{bar}(E_n))_{n\in\mathbb{N}} satisfies the Cauchy condition, thus the whole sequence admits a limit \bar \xi\in\mathbb{T}^N. Moreover, the convergence is exponentially fast in the sense that

for n large enough. Recalling (?) we thus conclude that there exists a suitable translation \xi \in \mathbb{T}^N such that for every k\in\mathbb{N}

Step 3. (Exponential convergence of the sets) By the previous step we can write, for n large, the boundaries of the evolving sets as radial graphs of the limit set E-\xi. Precisely, for n large enough there exist functions f_n such that

From (5.3) and for n large enough we have \|f_n-f_{n-1}\|_{L^2(\partial E)}\le 2\sqrt{\mathcal D(E_n,E_{n-1})} and thus, recalling (?) and arguing as in Step 2, we get

where b is as in (?). The above estimate yields the exponential decay of the L^2-norms of the radial graphs. We recall the well-known Gagliardo-Nieremberg inequality: for every j\in\mathbb{N} there exists C>0 such that, if g is smooth enough on the boundary of a smooth set E, then

where D^k stands for the collection of all the k-th order derivatives of g, see e.g. [Aub, Theorem 3.70]. Now, by (?) for every k there exists n_k such that \sup_{n\ge n_k}\| D^{2k} f_n\|_{L^2(\partial E)}\le 1, therefore we may apply (?) to f_n to deduce from (?) that also \| D^k f_n\|_{L^2(\partial E)} decays exponentially fast for all k\in\mathbb{N}. The Sobolev immersion Theorem then yields the exponential decay in C^k for every k thus completing the proof of the result.

First of all we notice that, if we take the constant C in (1.3) to be bigger than \sqrt{P(E)/2}, then it is enough to consider only the case \|H_{E_f}-\overline H_{E_f}\|_{L^2(\partial E)}\leq 1.

Set p=x+f(x)\nu_E(x) and let \varphi \in C^1(\partial E), by the definition of scalar mean curvature H_{E_f} and a change of coordinates we obtain

Combining (?), (?) and (?) we obtain

In the following, with a slight abuse of notation, with the symbol O(g) we will mean any function h of the form h(x)=r(x)g(x), where |r(x)|\le C for all x\in \partial E and C is a constant depending only on N and E.

By a simple Taylor expansion we have

From (?) and again by Taylor expansion, we obtain

where \nabla f,\nabla \varphi are respectively the tangent gradient of f,\varphi on \partial E and h is a vector field satisfying |h|\leq C(|f|+|\nabla f|^2))|\nabla f|. Set R=O(f^2)+O(|\nabla f|^2), by comparing (?) and (?) we infer that

Testing (?) with \varphi=1, we get

then, for \delta sufficiently small, using Hölder inequality we obtain

with C=C(N,E) since \delta\le 1. For \delta small enough, recalling that \|H_{E_f} - \overline H_{E_f}\|_{L^2}\leq1, the previous inequality implies

Using the bound \|f\|_{C^1}\leq \delta and the definition of h we easily see that

Testing (?) with \varphi=f, using Hölder’s inequality and by the previous remark, we get

By (3.2), (?) and by Hölder inequality, we obtain

Finally, by the above inequality, (3.2) again and by combining (?) with (?) we deduce that, for any \eta>0, it holds

The conclusion then follows combining (?) with Lemma 3.3 and taking \delta and \eta sufficiently small.

The first formula is a straightforward consequence of the area formula

and of the expression of the Jacobian J\Phi in (?). Now, (?) easily follows by taking the derivatives

in the first formula.

Set g=f-\bar f, where \bar f=\fint_{\partial E}f \,\textnormal{d} \mathcal{H}^{N-1}, then g has zero average and, by the first inequality in (3.3), we have

If \delta is sufficiently small, from (Proof) we obtain

Using the previous inequality, (Proof) again and the second inequality in (3.3) we infer that the function g satisfies

Then, we can apply Lemma 2.9 to obtain

provided \delta small enough. We conclude

up to taking \delta smaller if needed, and where the constant C>0 only depends on E.

The first statement is an immediate consequence of (?). Since E is a stationary point for the discrete flow, it satisfies

for all X \in C^1_c(\mathbb{T}^N,\mathbb{T}^N), i.e. E is a critical point for the perimeter.

The second part follows using the same argument of the proof of [MPS, Proposition 3.2]. Indeed, recall that the second variation has the following expression

which is positive if h is small enough. Then we procede as in the proof of [MPS, Proposition 3.2].

Analogously, we prove that E is the unique volume-constrained minimizer of \tilde J_h. Firstly, observe that, by Theorem 2.11, E is a strict local L^1-minimizer of the perimeter and it is a global minimizer of the second term in \tilde J_h. Therefore, there exists \varepsilon>0 such that

for all measurable set F such that |F|=|E| and |E \triangle F|\le \varepsilon, i.e. E is an isolated local minimizer for \tilde J_h in L^1 with the volume constraint, with minimality neighbourhood uniform with respect to h. Now, given any sequence (h_n)_{n \in \mathbb{N}} going to zero, let F_n be a volume constrained minimizer of J_{h_n}; we then easily deduce that |E \triangle F_n| \to 0 as n \to \infty, and therefore, for n large enough, |E\triangle F_n|\le \varepsilon. The strict minimality of E therefore implies F_n=E.

Let h^*=h^*(E) be the constant given by Proposition 4.3 so that, for every h< h^*, E is the unique volume-constrained global minimizer of the functional

Fix h<h^* and let (E_n)_{n \in \mathbb{N}} be a sequence of sets satisfying

Consider F_n a solution of (?) with E_n replacing E. We claim that

If we prove the claim, the conclusion easily follows.

First, Remark 4.1 ensures that (F_n)_{n \in \mathbb{N}} is a sequence of sets with uniformly bounded perimeters, with the bound depending only on N,m,h. Therefore, there exist F a set of finite perimeter such that |F|=m and a (unrelabelled) subsequence of (F_n)_{n \in \mathbb{N}} such that

Now, let K be a compact subset of \mathbb{T}^N such that, up to a subsequence, we have

From the second property in (Proof) we easily deduce that (\overline E)^c \subset K, and therefore K^c\subset \overline E. In particular, this inclusion implies that

for every G \subset \mathbb{T}^N. Setting

from the previous remark and from the fact that E is the unique minimizer of (Proof), we have

for any measurable set G\subset \mathbb{T}^N with |G|=|E|. Finally, we obtain

where we exploited the lower-semicontinuity of the perimeter and the minimality of F_n. Since E is the unique volume-constrained minimizer of \bar{J}_h, the set F must coincide with E and this concludes the proof.

Fix h< h^*, where h^*=h^*(E) is the constant given by Lemma 4.4 and let \sigma=\sigma(E), C=C(E) be the constants of Theorem 2.11. Moreover, let \delta:= \delta(\sigma,E) be the constant given by Lemma 4.4 with \sigma replacing \varepsilon. Set \delta^*\le \delta to be chosen later and consider E_0 such that

Recall that, from Remark 4.5 and from the hypothesis \overline{E}_0 \subset (E)_{\delta^*}, without loss of generality, we can assume \,\textnormal{d}_H(\overline{E}_0,\overline{E}) \le \delta^*. Moreover, by the regularity of E, we can also suppose \alpha(E_0,E)\le \tilde C\delta^*, for a suitable constant \tilde C>0 that only depends on E. From Lemma 4.4 we have that

Let x_0 be such that \alpha (E_0,E)= \lvert E_0 \triangle (E+x_0) \rvert. By choosing E+x_0 as a competitor for the minimality of E_h^1 and estimating dist_{\partial E_0}\leq \text{diam}(\mathbb{T}^N)= \sqrt{N}, we find

By (Proof), we can apply Theorem 2.11 and the previous estimate to obtain

where we have chosen \delta^* such that \delta^* \le C h\left (\min \{ \sigma, \delta,\varepsilon\} \right )^2/ (\tilde C\sqrt{N}). Since E_h^1 is a \Lambda-minimizer and E is regular, up to taking \delta^* smaller, the classical regularity theory for \Lambda-minimizers (see Theorem 2.4) implies

where x_1 is such that \alpha(E_h^1,E)=|E_h^1 \triangle (E+x_1)|.

Now we iterate the procedure: by induction, suppose that

where x_{n-1} is such that |E_h^{n-1} \triangle (E +x_{n-1})|=\alpha(E_h^{n-1},E). Observe that the second inequality in (?) implies that \,\textnormal{d}_H(\overline{E}_h^{n-1},\overline{E}+x_{n-1})\le \delta, therefore E_h^{n-1} and E+x_{n-1} satisfy the hypotheses of Lemma 4.4 and thus

Observe that by definition \alpha(E_h^n,E+x_{n-1})=\alpha(E_h^n,E). Now, by Theorem 2.11 and the monotonicity of the perimeters along the discrete flow we obtain

Again, thanks to the choice of \delta^*, the hypotheses of Theorem 2.4 are satisfied and thus

where x_n is such that \alpha(E_h^n,E)=|E_h^n\triangle (E+x_n)|. This concludes the proof.

Let m>0 such that the ball of volume m has perimeter strictly less than the one of the lamella of the same volume; we remark that for every smaller volume m'\le m the same property holds. Let E be a lamella of measure m, recall that E is a strictly stable set of the perimeter in \mathbb{T}^N. From the assumption on m it follows that E is only a local minimizer of the perimeter and not a global one.
Step 1. Firstly, we construct a sequence (E_n)_{n \in \mathbb{N}} such that E_n \to E in L^1 and \partial E_n \to \mathbb{T}^N in the Hausdorff distance. We define E_n by adding to E some balls contained in \mathbb T^N\setminus E and of overall small volume, and by subtracting to E balls contained in E with the same overall volume.

Recall that \mathbb{T}^N=[0,1]^N/\mathbb{Z}^N. In the following, with a little abuse of notation, we will identify \mathbb{T}^N and [0,1)^N. We define

for every n \in \mathbb{N}. Up to choosing m smaller, we can assume that m=1/2^s for some s \in \mathbb{N}. Moreover, we can suppose, up to translations, that E=[0,1)^{N-1} \times (0,1/2^s), thus for n \ge s we have

where \textnormal{Int}(\cdot) denotes the interior of a set in \mathbb{T}^N.

For every n\ge s and \underline k \in I_n, we consider the balls B_{n,\underline k} \subset Q_{n,\underline k} centered in the center of the cube Q_{n,\underline k} and of radius r_{n,\underline k} chosen in such a way that

Moreover, we can also take the radii r_{n,\underline k} sufficiently small so that

Set now

Define E_n=(E\cup C_n)\setminus A_n and observe that, by (?), we have |E_n|=|E|. Now, by (?), we also obtain

Observe that, from the definition of A_n and C_n, we have that

and therefore the set \partial E_n= \partial E \cup \partial C_n \cup \partial A_n converges in the Hausdorff metric to the whole \mathbb{T}^N as n\to +\infty. Therefore we have constructed a sequence (E_n)_{n \in \mathbb{N}} that satisfies

Step 2. Let E_n be the sets previously defined. We consider the space X=\{ F\subset \mathbb{T}^N\,:\, F \text{ is measurable} \} endowed with the L^1-distance, i.e. \textnormal{dist}_{L^1}(F,G)=|F\triangle G| for every F,G\in X. We extend our functional in the following way

and we set J_n:=\tilde J_h^{E_n}. We then prove the \Gamma-convergence of the functionals J_n to the perimeter functional in X, that is

We can clearly restrict ourselves to consider sets of finite perimeter and volume m, otherwise the result is trivial. For any given set F of measure m and finite perimeter we choose the sequence constantly equal to F as a recovery sequence for F. Indeed, by (?) we have

We now prove the \liminf inequality. Given a sequence F_n that converges to F in L^1, by the L^1-semicontinuity of the perimeter, we have

and thus (?) is proved.

Therefore, thanks to the equi-coercivity of the functionals J_n, any sequence of volume-constrained global minimizers of J_n converges in L^1, up to a subsequence, to a volume-constrained global minimizer of the perimeter in the torus. Let (F_n)_{n \in \mathbb{N}} be a sequence of global minimizers of the functional J_n and let F be such that F_n\to F in L^1. We know that F is a global minimizer of the perimeter and that by the choice of m the lamella is not a global minimizer. Therefore it must hold |E\triangle F|>0.

Let \varepsilon>0 be sufficiently small and let \delta^*=\delta^*(\varepsilon,E), h^*=h^*(E) be the constants given by Proposition 4.6. Fix E_0 an initial set satisfying |E|=|E_0| and \overline E_0\subset (E)_{\delta^*}. It is enough to show that any (unrelabelled) subsequence of the discrete flow starting from E_0 admits a further subsequence converging in C^k and up to translations to E. We divide the proof into three steps.
Step 1. (Existence and regularity of a limit point) From Proposition ? we remark that, for n \ge 1, the sets E_h^n are uniform \Lambda-minimizers with uniformly bounded, non-increasing perimeters. Therefore, by the compactness of (uniform) \Lambda-minimizers, we can conclude that there exists a subsequence (E_h^{k_n})_{n \in \mathbb{N}} and a \Lambda-minimizer E_h^{\infty} such that

Let G be a set of finite perimeter such that |G|=m. By the minimality of E_h^{k_n} we have

and, taking the limit as n\to \infty, we obtain

We have thus proved that E_h^{\infty} is a fixed point for the discrete flow and thus, by Proposition 4.3, it is a critical point for the perimeter.

Let \tau_\infty\in\text{argmin}_x|(E_h^\infty+x)\triangle E|. By Proposition 4.6 we have \alpha(E,E_h^{k_n}) \le \varepsilon for every n \in \mathbb{N}. Now, up to taking \varepsilon smaller, Theorem 2.4 and the smoothness of E, yields both the C^{1,\beta}-closeness between E_{h}^{\infty}+\tau_\infty and E, and the C^{1,\beta} regularity of E_h^\infty+\tau_\infty (and thus of E_h^\infty), for every \beta\in (0,1). From Proposition ? (iv) it follows that E_h^\infty is of class C^{2,\beta}, therefore we conclude that E_h^\infty has constant classical mean curvature and thus it is of class C^\infty. To conclude, the smoothness of E^\infty_h allow us to use Theorem 2.4 to improve the convergence of the subsequence to

and to ensure that the sets E^{k_n}_h are of class C^{1,\beta} for n large enough.
Step 2. (Convergence in C^{2,\beta} of the flow and C^{2,\beta}-closeness to E) In this step we we will prove that E^\infty_h is C^{2,\beta}-close to E and that the convergence of E^{k_n}_h to E^{\infty}_h is in C^{2,\beta}. Without loss of generality, we assume that \alpha(E,E^\infty_h)=|E\triangle E^\infty_h| so that the translation introduced by the previous step does not appear.

First of all we remark that, owing to the compactness of \partial E^\infty_h, it suffices to show that the result holds locally. By a compactness argument and the definition of convergence of sets in C^{1,\beta} (Definition 2.3), up to rotations and relabelling the coordinates, we can find a cylinder C=B'\times (-L,L), where B'\subset \mathbb{R}^{N-1} is a ball centred at the origin, and functions f_\infty,f_n\in C^{1,\beta}(B';(-L,L)) describing locally \partial E^\infty_h\cap C and \partial E^{k_n}_h\cap C respectively. We remark that the convergence (?) now reads as

We now prove that the curvatures H_{E^{k_n}_h} of the sequence E^{k_n}_h are converging in C^{0,\beta} to the curvature of E^\infty_h in the following sense

We will follow an argument used in Step 3 of the proof of [AFM, Theorem 4.3].

Since we described \partial E^{k_n}_h\cap C as a graph, the following formula for the curvature of \partial E_h^{k_n} holds

and an analogous formula holds for \partial E^\infty_h. From (?) and the Euler-Lagrange equation (?), by integrating on B', we then obtain

where we set y=x'/|x'| and integrated by parts in the last line. We can then exploit the convergence (?) and the formula (?) for the curvature of E^\infty_h to prove

Now, Lemma 5.1 ensures that sd_{E^{k_n-1}_h}\to sd_{E^\infty_h} uniformly and we can use the convergence (?) to obtain

since \partial E^\infty_h\cap C=\{(x',f_\infty(x')): x'\in B')\} by definition. Therefore we find

We then conclude that (?) converges to H_{E^\infty_h}\mathcal H^{N-1}(B') and thus it must hold

From (?), the previous result and the fact that the signed distance functions are all equi-lipschitz, we conclude that for any \beta\in (0,1), the sequence (H_{E^{k_n}_h}(\cdot,f_{k_n}(\cdot))) is bounded in C^{0,\beta}(B') and thus it converges uniformly to H_{E^\infty_h}(\cdot,f_{\infty}(\cdot)). This proves the convergence (?).

We remark that the previous result also hold if we describe the sets of the flow E_h^{k_n} as normal deformations of E^\infty_h, that is there exist functions \varphi_{k_n}: \partial E^\infty_h\to \mathbb{R} such that E_h^{k_n}=(E_h^{\infty})_{\varphi_{k_n}}. In this case the convergence (?) reads as

and this and Lemma 5.1 ensure that

Now, the convergence of the curvatures reads as

We can then apply directly [AFM, Lemma 7.2] to obtain that the subsequence E^{k_n}_h is converging to E^\infty_h in C^{2,\beta}.

To prove the C^{2,\beta}-closeness of the limit point we argue by contradiction. Assume that a sequence of limit points (E^{\infty,l}_{h})_{l\in\mathbb{N}} is converging in C^{1,\beta} to E but there exists \sigma>0 such that

for every l large enough. Again, we describe locally \partial E^{\infty,l}_h and \partial E as graphs of suitable functions f_{\infty,l},f : B'\to (-L,L) and we can repeat the same argument previously employed to prove that

This time the argument is simpler, since the limit points are stationary sets for the perimeter and thus their Euler-Lagrange equation is

Again, Lemma 7.2 in [AFM] yields the desired contradiction.
Step 3. (Uniqueness up to translations and C^k convergence) By the previous step we can find a suitable function \varphi_\infty\in C^{2,\beta}(\partial E) such that E^\infty_h=E_{\varphi_\infty}. Up to introducing a further translation given by Lemma 2.10, the hypotheses of Theorem 1.3 are satisfied and thus

since the set E^\infty_h is a stationary set for the perimeter. Therefore E^\infty_h is a translated of the set E.

A standard bootstrap method based on the elliptic regularity theory combined with the Euler-Lagrange equation (?) yields the convergence in C^k for every k\in\mathbb{N}.

The proof of equations (5.3) and (5.3) are quite analogous to the corresponding ones in [MPS]. We recall it for the sake of completeness and to highlight the minor differences between the two versions.

We start by observing that for any \eta'>0, if \delta is sufficiently small, then for every p_0\in\partial E_{f_2} the boundary of E_{f_2} in a small disc must be contained in a cone

We then divide the rest of the proof into two steps.
Step 1. If \delta is small enough , for every point p=\lambda p_0\in B_{2\delta}(p_0)\ (\lambda>0), we have that

Indeed the second inequality is trivial by definition, since p_0\in\partial E_{f_2}. Concerning the first one, set q\in\partial E_{f_2} such that \textnormal{dist}(p,\partial E_{f_2})=|p-q|, in particular |p-q|\le |p-p_0|\le 2\delta. From (?) we infer that q\in G and thus we have

where we used the explicit formula for the projection of a point on a cone. If p_0:=s+f_2(s)\nu_E(s)\in\partial E_{f_2} with s\in \partial E, we set

for an appropriate constant c such that the quantities defined are regular. We deduce that

Keeping the same notation and using the coarea formula (also recall (3.2)), we infer that

Recalling that for every s \in \partial E we have that J\Phi(s,\cdot)-1 is Lipschitz continuous with constant H_E, for \delta small enough and using (?), we get

from which the second inequality in (5.3) follows by taking \delta small enough. On the other hand, by (?) we also have

from which the first inequality in (5.3) follows by taking \eta' and \delta small enough.
Step 2. The inequalities (5.3) and (5.3) are now easy consequences. Indeed, by (?) we have that, for every p_1=(1+f_1(s))\nu_E(s)\in\partial E_{f_1}, it holds

Therefore (5.3) follows from (?) and (?), by taking \eta' and \delta smaller if needed and using the same change of coordinates used previously (recall that J\Phi and its inverse are estimated from above by 1+C\delta for a suitable constant C>0).

Finally, we prove (5.3). For \delta small enough, we can bound the Jacobian by 2 and therefore we obtain

and the conclusion follows from (5.3).

We first employ a compactness argument and then use Lemma 5.1 to conclude. We start by fixing some notation. We denote by

any disjoint union of l_B=4^{-1}\pi m^{-1}P_{\infty}^2 discs each having radius 2m/P_{\infty}; we denote by

the complement of any disjoint union of l_{B^c}=4^{-1}\pi (1-m)^{-1}P_{\infty}^2 discs, each of radius 2(1-m)/P_{\infty}; we denote by

any disjoint union of l_L{\textcolor{black}{\le\ }}P_{\infty}/2 lamellae blackhaving the same slope (blackand possibly having different masses). We remark that, for every fixed P_\infty and m, the following holds

blackThis is clear if we compare the families \mathscr E_B, \mathscr E_{B^c} and a union of lamellae having the same slope. Since, by Remark 6.2, there is a finite number of possible slopes for the lamellae, we conclude (Proof). From Remark 6.1 the discrete flow is eventually C^1-close to a limit point of the form \mathscr E_L, \mathscr E_B or \mathscr E_{B^c}. Assume now by contradiction that the flow does not converge to a fixed configuration. Then, without loss of generality, we can assume that for every 0<\varepsilon<i/3 there exist infinitely many indexes such that

Therefore we get

To reach the contradiction (compare (Proof)), it is enough to show that for every \varepsilon>0 there exists n_0=n_0(\varepsilon) such that for every n\ge n_0 it holds

Assume by contradiction the existence of a subsequence n_k along which the flow satisfies

Up to a further subsequence, E_{n_k}\to F, with F being a set of the form \mathscr E_B, \mathscr E_L or \mathscr E_{B^c}. But then Lemma 5.1 implies \textnormal{sd}_{E_{n_k-1}}\to \textnormal{sd}_F uniformly, which is clearly a contradiction.

blackFinally, we observe that in case iii) the number of connected component is given by \frac{P_{\infty}}{2|(p,q)|}, where we used the same notation of Remark 6.2. Thus, l=P_\infty/2 if and only if (p,q) is equal to (0,1) or to (1,0)

that means that the lamella is either vertical or horizontal.