• Aucun résultat trouvé

Decomposing X w 0

7. Construction of H 0

7.1. Decomposing X w 0

Here we decompose thewhite 0-tileXw0 into white trees.

Consider the white 1-tiles in Xw0. We assume in the next lemma that they are all connected at all 1-verticesvin theinteriorofXw0, anddisconnectedat all 1-vertices onC.

The resulting white connection graph may not be connected.

Lemma 7.6. The white connection graph in Xw0 as above has exactly one (white) cluster that intersects all sides (0-edges).

Proof. LetK be a (white) cluster inXw0 as above. Consider one componentB (in the standard topological sense) of Xw0\K. We call the set a:=∂B∩K a boundary arc ofK.

Claim 1. Every boundary arc aas above is contained in a single black 1-tile.

Clearlya is a union of 1-edges. Eithera starts and ends at two distinct 1-vertices v, w∈C, or a is a closed curve. Let E, E03v be two 1-edges contained in a which are consecutive in∂B, wherev /∈C is a 1-vertex. Note that by construction all white 1-tiles Xj3v are connected atv. ThusE and E0 are contained in the same black 1-tile. The claim follows.

Assume now thatais a closed curve. Thenais a Jordan curve in the boundary of a single black 1-tile. Thus the corresponding componentB is the interior of a single black 1-tile. Henceadoes not separateK from any other distinct white clusterK0 inXw0.

We call a black 1-tileY⊂Xw0 non-trivial ifY∩Ccontains at least two 1-vertices. A complementary component ofY is the closure of a componentXw0\Y.

Claim2. LetX, X0⊂Xw0 be two distinct white 1-tiles. ThenXandX0are contained in distinct white clusters K, K0⊂Xw0 if and only if there is a black 1-tile Y⊂Xw0 such thatX andX0 are contained in complementary components ofY.

The implication (⇐) is clear. To see the other implication we note that if X0 is contained in a cluster distinct from the cluster K⊃X, then X0 has to be contained in the closure of one component ofXw0\K. Such a component is separated from K by a boundary arca. However, if adoes not contain two 1-vertices v, w∈C, this component is a single black 1-tile, meaning that it does not containX0. OtherwiseX0 is separated fromX by the black 1-tileY containinga, proving the claim.

Recall that we assumed that no 1-tile joins opposite sides ofC(see Lemma7.5). Thus for every non-trivial black 1-tile Y there is a complementary component ofY, denoted byKY, that intersects all 0-edges.

We now defineK:=T

Y KY, where the intersection is taken over all non-trivial black 1-tilesY⊂Xw0. Since two non-trivial black 1-tilesY, Y0⊂Xw0 do not cross, it follows that Kintersects all 0-edges.

By Claim 2 it follows that all white 1-tiles contained inKare connected, i.e., belong to the same cluster denoted byK.

Assume thatKintersects a given 0-edgeE0in a 1-edgeE. This cannot happen ifE is contained in a black 1-tileY⊂Xw0, sinceY would be non-trivial, and the corresponding setKY does not containE. ThusE is contained in a white 1-tile, which is inK.

IfKintersectsE0only in a 1-vertexv, there is a boundary arca⊂∂K containingv.

LetY⊂Xw0 be the corresponding non-trivial black 1-tile containing a. LetE⊂abe the 1-edge containingv. SinceE is not inC, the white 1-tile containingE is in K.

This means that there is a white 1-tile in Kthat intersectsE0.

In each white cluster inXw0 define a spanning tree (see Definition6.11). The spanning tree in the cluster from Lemma 7.6 is called the main tree KM, the spanning trees in the other clusters are called thesecondary trees inXw0. Theconnections at all 1-vertices v∈Xw0\C are thusdefined, they will not be changed any more in the construction.

Let E be the boundary circuit of the main treeKM (see Definition6.21 and Lem-ma6.23). Letv0, ..., vN−1be the 1-vertices onC thatE visits (in this order). Note that a 1-vertexv may appear several times in this list.

Notation. Given points v, w∈C, denote by [v, w] (resp. (v, w)), the closed (resp.

open) positively oriented arc onC fromv tow. Note that (v, v)=∅.

Lemma 7.7. The points {vi}Ni=0−1 satisfy the following conditions (indices are taken modN here):

(1) Each (open) arc (vi, vi+1) contains no point vl. This means that the points {vi}N−1i=0 are positively oriented on C.

(2) The points vi and vi+1 are not contained in disjoint 0-edges,in particular each 0-edge contains at least one point vi.

(3) For all i there is a black 1-tile Y3vi, vi+1.

(4) Let K be a secondary tree in Xw0. Then there is an arc [vi, vi+1]such that K∩C ⊂[vi, vi+1].

Proof. (1) LetKM,εbe a geometric representation ofKM as in Lemma6.23(3). The pathγionE betweenvi andvi+1 is then represented by a Jordan arcγi,εwith endpoints vi,ε andvi+1,εsuch that|vi−vi,ε|and|vi+1−vi+1,ε|are arbitrarily small. Since all white 1-tiles are disconnected at every 1-vertexv∈C, we may assume thatvi,ε∈Candγi,ε⊂Xw0 for alli.

The arcsγi,ε are non-crossing, and thus the points{vi,ε}Ni=0−1 are ordered cyclically or anti-cyclically onC. Hence the points{vi}N−1i=0 are ordered cyclically or anti-cyclically onC.

The winding number ofE aroundx /∈Eis 1 if and only ifxis in the interior of a white 1-tile of the main tree. This follows from an inductive argument as in Corollary6.20.

Assume that the points {vi}Ni=0−1 are ordered anti-cyclically on C. Let Ci be the (positively oriented) arc onC between vi and vi+1. Then γi+Ci has winding number 0 around any pointx in the interior of a 1-tile of the main tree. ThusE+C has winding number 0 around such anx. This is a contradiction.

(3) Considervi andvi+1. Then

• eithervi=vi+1 in which case the statement is trivial;

• or [vi, vi+1] is a 1-edge, property (3) is then clear again;

• orvi andvi+1 are the boundary points of a boundary arcaofKM, as in Claim 1 from the proof of Lemma7.6. In this case there is a black 1-tile Y⊃a.

(2) This follows immediately from (3) and the assumption that no 1-tile intersects disjoint 0-edges. Furthermore KM intersects a 0-edge E if and only if it intersects it in some 1-vertex. The set of all 1-vertices in which KM intersects C is equal to the set {vi}Ni=0−1. Thus, sinceKM intersects each 0-edge, it follows that each 0-edge contains one pointvi.

(4) The reader is reminded of Claims 1 and 2 in the proof of Lemma7.6. For every secondary componentKthere is an arcacontained in a (non-trivial) black 1-tileY such that intKis in the component ofXw0\anot intersecting all 0-edges. Letvi andvi+1be the endpoints ofa(see the discussion from (3)), then

K∩C ⊂[vi, vi+1].

Documents relatifs