§ 1. Hyperplanes, chambers and facets
In this section, E denotes a real affine space of finite dimension d and T the space of translations of E (cf. Algebra, Chap. II, § 9). If a and b are two points of E, [ab] (resp. ]ab[, resp. ]ab]) will denote the closed segment (resp. open segment, resp. segment open at a and closed at b) with extremities a, b. The space T is provided with its unique separated topological vector space topology, cf. Topological Vector Spaces, Chap. I, § 2, no. 3; it is isomorphic to . The space E is provided with the unique topology such that, for all , the map from T to E is a homeomorphism.
We denote by H a locally finite set of hyperplanes of E (General Topology, Chap. I, § 1, no. 5).
1. Notations
Let H be a hyperplane of E. Recall that E – H has two connected components, called the open half-spaces bounded by H. Their closures are called the closed half-spaces bounded by H. Let . Then x and y are said to be strictly on the same side of H if they are contained in the same open half-space bounded by H, or equivalently, if the closed segment with extremities x and y does not meet H; x and y are said to be on opposite sides of H if x belongs to one of the open half-spaces bounded by H and y to the other. If and , then x and t are said to be strictly on the same side of H if this is so for x and for all .
Let A be a non-empty connected subset of E. For any hyperplane H of E that does not meet A, denotes the unique open half-space bounded by H that contains A. If is a set of hyperplanes of E, none of which meet A, put
2. Facets
The set of points of E that do not belong to any hyperplane H of the set is open since is locally finite. More precisely, we have the following result:
Proposition 1
Let a be a point of E. There exists a connected open neighbourhood of a that does not meet any hyperplane H that belongs to and does not pass through a. Moreover, there exist only finitely many hyperplanes that belong to and pass through a.
The set of hyperplanes H such that and is locally finite since it is contained in . Hence, the set U of points of E that do not belong to any of the hyperplanes of the set is open. Since , there is a connected open neighbourhood of a contained in U. The remainder of the proposition is clear.
Given two points x and y of E, denote by R\{x, y\} the relation
" For any hyperplane , either and or and are strictly on the same side of H."
Clearly, R is an equivalence relation on E.
Definition 1
A facet of E relative to is an equivalence class of the equivalence relation defined above.
Proposition 2
The set of facets is locally finite.
This is clear since is locally finite.
Let F be a facet and a a point of F. A hyperplane contains F if and only if ; the set of these hyperplanes is thus finite; their intersection is an affine subspace L of E, which we shall call the affine support of F; the dimension of L will be called the dimension of F.
If is the set of hyperplanes not containing F, then
We shall prove that the closure of F is given by
It is clear that the right-hand side contains the left-hand side. Conversely, let . The open segment with extremities a and x is contained in L and in each of the for , and hence in F. It follows that x is in the closure of F, hence the formula.
Proposition 3
Let F be a facet and L its affine support.
(i) The set F is a convex open subset of the affine subspace L of E. (ii) The closure of F is the union of F and facets of dimension strictly smaller than that of F. (iii) In the topological space L, the set F is the interior of its closure.
Since every open half-space and every hyperplane are convex subsets of E, formula (2) shows that F is the intersection of a family of convex subsets, and hence is convex. On the other hand, let a be in F, and let U be a convex open neighbourhood of a in E that does not meet any of the hyperplanes in the set of such that . For any , we thus have , hence , so that F is open in the topological space L.
Let b be a point of , belonging to a facet , and let be the set of hyperplanes passing through b; put . For any H in , we have and , hence and ; by the definition of a facet, we thus have
whereas (3) implies that
hence . We cannot have , for this would imply that by (2) and (4), contrary to the hypothesis that and . The support of is the set ; we have , but for in , so and finally , that is . This proves (ii).
Let H be in and let D be the open half-space bounded by H and distinct from ; we have , and it is immediate that is a half-space of L bounded by the hyperplane of L; consequently, every neighbourhood of b in L meets , and since is disjoint from by (3), we see that the point b of cannot be in the interior of in the topological space L. Since F is open in L, we have (iii). Q.E.D.
Corollary
Let F and F' be two facets. If , the facets F and F' are equal.
This follows from (iii).
Proposition 4
Let F be a facet, and let L be an affine subspace of E that is the intersection of hyperplanes belonging to ; denote by the set of hyperplanes that do not contain L. The following conditions are equivalent:
(i) There exists a facet F' with support L that meets . (ii) There exists a facet F' with support L contained in . (iii) There exists a point x in that does not belong to any of the hyperplanes of .
If these conditions are satisfied, is the unique facet with support contained in .
(i) (ii): Since is a union of facets (Prop. 3, (ii)), every facet that meets meets a facet contained in , and so is equal to it.
(ii) (iii): Every point of satisfies (iii) since every hyperplane of containing contains , and hence .
(iii) (i): Let be a point satisfying (iii) and let be the facet containing ; it is clear that meets . Let ; then if and clearly if ; consequently the support of is the intersection of the hyperplanes of , and is equal to .
Finally, let be a facet with support contained in , and let be a point of ; since no hyperplane of passes through , Prop. 1 shows that there exists a convex open neighbourhood of that does not meet any hyperplane of . Since is in the closure of , ; now is the set of hyperplanes that do not contain , and for all in we have and formula (2) implies that
Q.E.D.
3. Chambers
Definition 2
A chamber of relative to (or simply a chamber if there is no ambiguity regarding ) is a facet of relative to that is not contained in any hyperplane belonging to .
Let be the open set in consisting of the points that do not belong to any hyperplane of ; since a hyperplane of must contain any facet that it meets, the chambers are the facets contained in ; every chamber is a convex (hence connected) open subset of by Prop. 3, (i); since the chambers form a partition of , they are exactly the connected components of . Every convex subset of is connected, and thus contained in a chamber, which is unique if is non-empty. It is clear that the chambers are the facets with support , and Prop. 3, (iii) shows that every chamber is the interior of its closure. Finally, let be a chamber and a non-empty subset of ; formulas (2) and (3) imply that
since for all .
Proposition 5
Let be a non-empty subset of . Assume that there exists a subset of with the following properties:
a) For any , there exists an open half-space bounded by such that .
b) *The set C does not meet any hyperplane belonging to . Under these conditions, C is a chamber defined by in E, and for all .
Properties a) and b) show that C is a convex subset of U; hence there is a chamber with . Since , we have for all H in , hence since ; we have by (6), hence . Finally therefore, .
Proposition 6
Every point of E is in the closure of at least one chamber.
If E reduces to a single point, this is clear. Otherwise, let and let be the hyperplanes of containing . Since is locally finite, there exists a neighbourhood V of that does not meet any hyperplane of other than . Let D be a straight line passing through and not contained in any of the ; if , and is sufficiently close to , the open segment is contained in V and does not meet any of the . Then ; since is connected, it is contained in a chamber C, hence .
Proposition 7
Let L be an affine subspace of E and a non-empty open subset of L.
(i) There exists a point a in that does not belong to any of the hyperplanes of that do not contain L.
(ii) If L is a hyperplane and , there exists a chamber that meets .
(iii) If L is a hyperplane and , there exists a point a in that does not belong to any hyperplane of .
Denote by the set of hyperplanes H with and , and by the set of hyperplanes of the affine space L of the form with . It is clear that is a locally finite set of hyperplanes in L, and Prop. 6 shows that meets a chamber defined by in L. If is a point of , then for all , hence (i).
Assume now that L is a hyperplane; any hyperplane containing L is then equal to it, so we may distinguish two cases:
a) : then , and we have for all ; thus belongs to a chamber defined by in E, hence (ii).
b) : then , hence (iii).
4. Walls and faces
Definition 3
Let C be a chamber of E. A face of C is a facet contained in the closure of C whose support is a hyperplane. A wall of C is a hyperplane that is the support of a face of C.
Every wall of C belongs to . Prop. 4 shows that a hyperplane is a wall of C if and only if . Moreover, every wall of C is the support of a single face of C.
Proposition 8
Every hyperplane* belonging to is the wall of at least one chamber.
By Prop. 7, (iii), there exists a point of that does not belong to any hyperplane of ; by Prop. 6, there exists a chamber such that ; Prop. 4 then shows that is a wall of .
Proposition 9
Let* be a chamber and the set of walls of . Then and every subset of such that contains . A subset of is a facet if and only if it is a facet of relative to the family .
a) Let be a subset of such that . Consider a hyperplane belonging to but not to ; let be the set of hyperplanes belonging to . Then , hence , and does not meet . By the implication (i) (iii) in Prop. 4, the hyperplane is not a wall of . Consequently, every wall of belongs to .
b) We assume that . Let be a hyperplane belonging to that is not a wall of , and put . By the implication (iii) (i) in Prop. 4, the convex set does not meet , so and . If is a finite subset of that does not contain any wall of , we conclude by induction on the cardinal of that .
c) Let be a point of ; clearly, . Let be a point of ; since the closed segment is compact, the set of hyperplanes that meet is finite. Since and are strictly on the same side of every wall of , no wall of belongs to ; by b), we have . Since , we have . We have therefore proved that , which establishes the first part of the proposition.
d) To prove the last assertion of the proposition, it clearly suffices to show that a subset of that is a facet of relative to is a facet of relative to , or that every hyperplane that meets contains . So let be a hyperplane that meets but does not contain it. Since is open in its affine support, it is not completely on one side of . It follows that is not completely on one side of and hence that the hyperplane does not belong to , which completes the proof.
Remark 1
It follows from formula (6) and Prop. 9 that the closure of a chamber is the intersection of the closed half-spaces that are bounded by a wall of and contain .
Remark 2
Let be a facet whose support is a hyperplane ; we shall show that is a face of two chambers. Let be the set of hyperplanes belonging to ; put and denote by and the two open half-spaces bounded by . The set is open and contains , and since every point of is in the closure of and , the sets and are non-empty; these are chambers. Moreover, the hyperplane meets ; Prop. 4 shows that is a wall of and that
F, which meets , is a face of ; similarly, F is a face of . Finally, let C be a chamber of which F is a face, and suppose for example that ; by Prop. 4, the set meets F, and hence is equal to , and we have
5. Intersecting hyperplanes
Recall (Algebra, Chap. II, § 9, no. 3) that two affine subspaces P and of E are said to be parallel if there exists a vector t in T such that . It is clear that the relation "P and are parallel" is an equivalence relation on the set of affine subspaces of E.
Lemma 1
Any two non-parallel hyperplanes have non-empty intersection.
Let H and be two non-parallel hyperplanes, and ; there exist two hyperplanes M and of the vector space T such that and ; since H and are not parallel, we have and hence ; hence there exist and such that , and the point belongs to .
Lemma 2
Let H and be two distinct hyperplanes of E, and f, two affine functions on E such that H (resp. ) consists of the points a in E such that (resp. ). Finally, let L be a hyperplane of E. Assume that one of the following hypotheses is satisfied: a) The hyperplanes H, and L are parallel. b) The hyperplanes H and are not parallel, and . Then there exist real numbers , not both zero, such that L consists of those points at which the affine function vanishes.
The lemma being trivial when , we assume that there exists a point in L with . Put and
then since ; moreover, since , there exists such that , so that , and thus is non-zero. The set of points where the affine function vanishes is then a hyperplane of E; we have , so .
a) Assume that H and are parallel: g and f both vanish at every point of , hence so does since ; thus every point of belongs to ; but since H and are parallel and distinct, they are disjoint, so , and Lemma 1 shows that is parallel to H. Since and , we thus have .
b) Assume that H and are not parallel: by Lemma 1, there is a point c in ; we give E the vector space structure obtained by taking c as the origin. Then is a vector subspace of of codimension 2, and since , the vector subspace of generated by and is a hyperplane; since and , we have , hence . Q.E.D.
Proposition 10
Let C be a chamber, let H and H' be two walls of C, and let L be a hyperplane meeting . Assume that H is distinct from H' and that one of the following conditions is satisfied:
a) The hyperplanes H, H' and L are parallel.
b) The hyperplanes H and H' are not parallel, and .
Then, L meets C.
Let (resp. ) be a point of the face of C with support H (resp. ); it is immediate that every point of the segment distinct from and belongs to C.
Introduce an affine function that vanishes at every point of H and is such that for in ; similarly, introduce an affine function having an analogous property with respect to . By applying Lemma 2, we can find numbers and and an affine function having the properties stated in the lemma. We have , and for every point of , we have and , so . On the other hand, we have and , and since , we have . The points and are thus strictly on opposite sides of the hyperplane L, hence there exists a point c of L which belongs to , and is distinct from and , hence that belongs to C.
6. Examples: simplicial cones and simplices
a) Let be a point of , and a basis of ; every point of can thus be written uniquely in the form
with being real numbers. Denote by the affine function on which, at any , written in the form (7), takes the value . Denote by the hyperplane formed by those such that , and by the set of hyperplanes . For every subset J of the set , put ; for every sequence of numbers equal to 0, 1 or -1, denote by the set of those such that is of sign for all in I (General Topology, Chap. 4, § 3, no. 2). It is immediate that the facets defined by in are the sets and that these sets are pairwise distinct; the support of is equal to if J is the set of such that ; in particular, the chambers are the sets of the form where each of the numbers is equal to 1 or -1.
The set formed by those with for all is a chamber, called the open simplicial cone with vertex a defined by the basis (e_1, \ldots, e_d). Its closure consists of the points x such that e_i'(x) \geq 0 for all i \in I when d \geq 1; otherwise, its closure is empty. For every subset J of I, let C_J be the set of points x of E such that e_i'(x) = 0 for i \in J and e_i'(x) > 0 for i \in I - J. Then C_J is a facet with support H_J, and it is an open simplicial cone with vertex a in the affine space H_J; moreover, . In particular, the walls of C are the hyperplanes H_i for i \in I, and the face of C contained in H_i is equal to C_{\{i\}}.
None of the sets H_i, H_J, C, C_J and F(\varepsilon_1, \ldots, \varepsilon_d) change if we replace the basis (e_1, \ldots, e_d) by a basis (\lambda_1 e_1, \ldots, \lambda_d e_d) with \lambda_i > 0 for all i.
b) Suppose now that we are given an affinely free system of points of E, say (a_0, a_1, \ldots, a_d). We know that every point of E can be written uniquely in the form , where \xi_0, \ldots, \xi_d are real numbers with \xi_0 + \cdots + \xi_d = 1 (Algebra, Chap. 2, § 9, no. 3). Define affine functions , the function associating to the point x, written as above, the number \xi_i. Denote by H_i the hyperplane of E defined by and by the set of hyperplanes H_0, H_1, \ldots, H_d; finally, put I = \{0, 1, \ldots, d\}. The open simplex with vertices a_0, \ldots, a_d is the set C of points x of E such that for all i \in I; it is one of the chambers defined by in E. The closure of C is the set of points x such that for all i \in I; it is the convex envelope of the finite set and it is easy to see that the extreme points of are .
For any subset J of I distinct from I, put and let C_J be the set of points x of E such that for i \in J and for i \in I - J. The set C_J is an open simplex in the affine space H_J with vertices the points a_i for i \in I - J; we have and if ; moreover, C_J is a facet with support H_J. In particular, the walls of C are H_0, \ldots, H_d and C_{\{i\}} is the face contained in H_i.
For any non-empty subset K of I, let B_K be the set of points x of E such that for i \in K and for i \in I - K. The sets B_K are the chambers defined by in E and we have B_I = C. It is easy to see that is compact; on the other hand, if K is a subset of I distinct from I, of cardinal p, the chamber B_K contains the sequence of points defined for n \geq 2 by
showing that B_K is not relatively compact.