Boundaries of systolic groups
Damian Osajda, Piotr Przytycki

TL;DR
This paper constructs EZ-structure boundaries for systolic groups, leading to the proof of the Novikov conjecture for torsion-free cases, by analyzing geodesic-like structures in systolic complexes.
Contribution
It introduces a novel boundary construction for systolic groups that parallels CAT(0) geodesics, enabling new topological and conjectural results.
Findings
Boundaries are EZ-structures for all systolic groups
Proves the Novikov conjecture for torsion-free systolic groups
Establishes properties of systolic geodesics similar to CAT(0) spaces
Abstract
For all systolic groups we construct boundaries which are EZ--structures. This implies the Novikov conjecture for torsion--free systolic groups. The boundary is constructed via a system of distinguished geodesics in a systolic complex, which we prove to have coarsely similar properties to geodesics in CAT(0) spaces.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsGeometric and Algebraic Topology · Topological and Geometric Data Analysis · Mathematical Dynamics and Fractals
Boundaries of systolic groups
Damian Osajda 111Partially supported by MNiSW grant N201 012 32/0718. This research was supported by a Marie Curie European Reintegration Grant within the 6th European Community Framework Programme.
Instytut Matematyczny, Uniwersytet Wrocławski, pl. Grunwaldzki 2/4, 50–384 Wrocław, Poland
Piotr Przytycki 222Partially supported by MNiSW grant N201 003 32/0070, MNiSW grant N201 012 32/0718, and the Foundation for Polish Science.
Institute of Mathematics, Polish Academy of Sciences, Śniadeckich 8, 00-956 Warsaw, Poland
Abstract
For all systolic groups we construct boundaries which are –structures. This implies the Novikov conjecture for torsion–free systolic groups. The boundary is constructed via a system of distinguished geodesics in a systolic complex, which we prove to have coarsely similar properties to geodesics in spaces.
MSC: 20F65; 20F67; 20F69;
Keywords: Systolic group, simplicial nonpositive curvature, boundaries of groups, –set compactification
Contents
- 1 Introduction
- 2 Systolic complexes
- 3 Definition of the boundary
- 4 Topology on
- 5 Compactness and finite dimensionality
- 6 The main result
- 7 Flat surfaces
- 8 Layers
- 9 Euclidean geodesics
- 10 Directed geodesics between simplices of Euclidean geodesics
- 11 Euclidean geodesics between simplices of Euclidean geodesics
- 12 Characteristic discs spanned on Euclidean geodesics
- 13 Contracting
- 14 Final remarks
1 Introduction
There are many notions of boundaries of groups used for various purposes. In this paper we focus on the notions of –structure and –structure introduced by Bestvina [Be] and studied e.g. in [D], [FL]. Our main result is the following.
Theorem A** (Theorem 6.3).**
Let a group act geometrically by simplicial automorphisms on a systolic complex . Then there exists a compactification of satisfying the following:
* is a Euclidean retract (ER),* 2. 2.
* is a –set in ,* 3. 3.
for every compact set , is a null sequence, 4. 4.
the action of on extends to an action, by homeomorphisms, of on .
A group as in Theorem A is called a systolic group. It is a group acting geometrically (i.e. cocompactly and properly discontinuously) by simplicial automorphisms on a systolic complex—contractible simplicial complex satisfying some local combinatorial conditions. Systolic complexes were introduced by Januszkiewicz–Światkowski [JS] and, independently, by Haglund [H] and by Chepoi [C] (in Section 2 we give some background on them). Systolic complexes (groups) have many properties of non–positively curved spaces (groups). There are systolic complexes that are not when equipped with the path metric in which every simplex is isometric to the standard Euclidean simplex. On the other hand, there are systolic groups that are not hyperbolic, e.g. . Summarizing, systolic setting does not reduce to the or to the hyperbolic one — it turns out that systolic groups form a large family: allow various combinatorial constructions [H],[JS],[JS+] and provide the discipline with new range of examples frequently with unexpected properties [JS2],[JS+]. We also believe that eventually both systolic complexes and cubical ones will be placed among a wider family of combinatorially non–positively curved contractible cell complexes.
Here we give the other definitions that appear in the statement of Theorem A. A compact space is a Euclidean retract (or ER) if it can be embedded in some Euclidean space as its retract. A closed subset of a Euclidean retract is called a –set if for every open set , the inclusion is a homotopy equivalence. A sequence of subsets of a topological space is called a null sequence if for every open cover of all but finitely many are –small, i.e. for all but finitely many there exist such that .
Conditions 1, 2 and 3 of Theorem A mean (following [Be], where only free actions are considered, and [D]) that any systolic group admits a –structure . The notion of an –structure, i.e. a –structure with additional property 4 was explored by Farrell–Lafont [FL] (in the case of a free action).
Bestvina [Be] showed that some local homological invariants of the boundary are related to cohomological invariants of the group. In particular, the dimension of the boundary is an invariant of the group i.e. it does not depend on the –structure we choose. This was generalized by Dranishnikov [D] to the case of geometric actions. It should be emphasized that the homeomorphism type of the boundary is not a group invariant (but the shape is an invariant [Be]), as the Croke–Kleiner examples of visual boundaries of some spaces exhibit [CrK]. Carlsson–Pedersen [CP] and Farrell–Lafont [FL] proved that existence of an –structure on a torsion–free group implies that the Novikov conjecture is true for . Thus, by Theorem A, we get the following.
Corollary**.**
Torsion–free systolic groups satisfy the Novikov conjecture.
There are only few classes of groups for which a –structure has been found (and even fewer for which an –structure is known). The most important examples are: hyperbolic groups [BeMe] —with being the Rips complex and being the Gromov boundary of ; groups —with a space and the visual boundary of ; relatively hyperbolic groups whose parabolic subgroups admit a –structure [Da]. Bestvina [Be] asked whether every group with finite has a –structure.
The question whether for every systolic group there exists an –structure was posed by Januszkiewicz and Światkowski in 2004. Theorem A answers affirmatively this question.
We hope that, similarly to the hyperbolic and cases, our boundaries will be also useful for purposes other than the ones mentioned above. In particular we think that splittings of systolic groups can be recognized through the topology of the boundary, as in e.g. [B], [PS]. Studying more refined structures on the boundary could help in obtaining rigidity results for some systolic groups.
The essential point of our construction is the choice of the system of good geodesics (derived from the system of Euclidean geodesics, the distinction being not important at this moment), which is coarsely closed under taking subsegments (Theorem B below), and which satisfies coarsely a weak form of condition (Theorem C below).
Recall that Januszkiewicz and Światkowski [JS] considered a system of directed geodesics in a systolic complex (c.f. Definition 2.11). One may try to define the boundary of a systolic complex by taking the inverse limit of the following system. Consider the sequence of combinatorial spheres around a fixed vertex and projections from larger to smaller spheres along the directed geodesics terminating at . Unfortunately, the inverse limit of this system does not have, in general, property 3 from Theorem A. Property 3 fails, for example, already for the flat systolic plane (c.f. Definition 7.1).
Hence, instead of using directed geodesics, we introduce Euclidean geodesics, which behave like geodesics with respect to the flat subcomplexes a systolic complex. To define the Euclidean geodesic between two vertices, say , in a systolic complex, we consider the loop obtained by concatenating the two directed geodesics joining to and to . Then we span a minimal surface on this loop. (Here Elsner‘s minimal surfaces theory [E] comes in handy. To obtain some uniqueness properties on we complement Elsner‘s theory with our results on layers, which span the union of all 1–skeleton geodesics between and .) The surface is isometric to a contractible subcomplex of the flat systolic plane and hence has a natural structure of a space. The Euclidean geodesic is defined as a sequence of simplices in , which runs near the geodesic between and .
Now we pass to the more technical part of the exposition. Formally, the Euclidean geodesic is defined for a pair of simplices in a systolic complex satisfying for some (where denotes the combinatorial sphere of radius around , c.f. Definition 2.4). The Euclidean geodesic is a certain sequence of simplices , where , such that , which satisfies for (c.f. Lemma 9.15(i)). The two most significant features of Euclidean geodesics are given by the following.
Theorem B** (Theorem 12.2).**
Let be simplices of a systolic complex , such that for some natural we have . Let be the Euclidean geodesic between and . Take some and let be a 1–skeleton geodesic such that for . Consider the simplices of the Euclidean geodesic between vertices and . Then for each we have , where is a universal constant.
Theorem C** (Theorem 13.1).**
Let be vertices in a systolic complex such that . Let be 1–skeleton geodesics such that , where are Euclidean geodesics for and for respectively. Then for all we have , where is a universal constant.
The article is organized as follows. It consists of an introductory part (Sections 1–2), the two main parts (Sections 3–6 and Sections 7–13), which can be read independently, and of a concluding Section 14.
In Section 2 we give a brief introduction to systolic complexes.
In the first part, assuming we have defined Euclidean geodesics satisfying Theorem B and Theorem C, we define the boundary: In Section 3 we define the boundary as a set of equivalence classes of good geodesic rays. Then we define topology on the compactification obtained by adjoining the boundary (Section 4) and we show its compactness and finite dimensionality (Section 5). Finally, in Section 6, we prove Theorem A —the main result of the paper.
In the second part of the article we define Euclidean geodesics and establish Theorem B and Theorem C: In Section 7 we recall Elsner‘s results on minimal surfaces. In Section 8 we study layers, whose union contains all geodesics between given vertices. We define Euclidean geodesics in Section 9.
In the next two sections we prove Theorem 10.1 which is a weak version of Theorem B (though with a better constant). Apart from the definitions these sections can be skipped by a hurried reader. We decided to include them since this way of obtaining (the weak version of) Theorem B is straightforward in opposition to the strategy in Section 12, which is designed to obtain Theorem C. In Section 10 we study the position of directed geodesics between two simplices of a given Euclidean geodesic with respect to the minimal surface appearing in its construction. Then we verify Theorem 10.1 in Section 11 by studying geometry of minimal surfaces.
The last two sections are devoted to the proofs of Theorem B and Theorem C: In Section 12 we prove (in a technically cumbersome manner) powerful Proposition 12.1 linked with properties of the triangles, whose two sides are Euclidean geodesics. Proposition 12.1 easily implies Theorem B, but its main application comes in Section 13, where we use it to derive Theorem C.
We conclude with announcing some further results for which we do not provide proofs in Section 14.
Acknowledgments. We are grateful to Tadeusz Januszkiewicz and Jacek Światkowski for discussions right from the birth of our ideas and to Mladen Bestvina for encouragement. We thank the Mathematical Sciences Research Institute and the Institut des Hautes Études Scientifiques for the hospitality during the preparation of this article.
2 Systolic complexes
In this section we recall (from [JS],[JS2],[SH]) the definition and basic properties of systolic complexes and groups.
Definition 2.1**.**
A subcomplex of a simplicial complex is called full in if any simplex of spanned by vertices of is a simplex of . The span of a subcomplex is the smallest full subcomplex of containing . We denote it by span. A simplicial complex is called flag if any set of vertices, which are pairwise connected by edges of , spans a simplex in . A simplicial complex is called –large, , if is flag and there are no embedded cycles of length , which are full subcomplexes of (i.e. is flag and every simplicial loop of length and ”has a diagonal”).
Definition 2.2**.**
A simplicial complex is called systolic if it is connected, simply connected and links of all simplices in are 6–large. A group is called systolic if it acts cocompactly and properly by simplicial automorphisms on a systolic complex . (Properly means is locally finite and for each compact subcomplex the set of such that is finite.)
Recall [JS], Proposition 1.4, that systolic complexes are themselves 6–large. In particular they are flag. Moreover, we have the following.
Theorem 2.3** ([JS], Theorem 4.1(1)).**
Systolic complexes are contractible.
Now we briefly treat the definitions and facts concerning convexity.
Definition 2.4**.**
For every pair of subcomplexes (usually vertices) in a simplicial complex denote by ( for vertices ) the combinatorial distance between in , the 1–skeleton of (i.e. the minimal number of edges in a simplicial path connecting both sets). A subcomplex of a simplicial complex is called 3–convex if it is a full subcomplex of and for every pair of edges such that , we have . A subcomplex of a systolic complex is called convex if it is connected and links of all simplices in are 3–convex subcomplexes of links of those simplices in .
In Lemma 7.2 of [JS] authors conclude that convex subcomplexes of a systolic complex are full and 3–convex in , and systolic themselves, hence contractible by Theorem 2.3. The intersection of a family of convex subcomplexes is convex. For a subcomplex , , the combinatorial ball of radius around is the span of . (Similarly .) If is convex (in particular, if is a simplex) then is also convex, as proved in [JS], Corollary 7.5. Combining this with previous remarks we record:
Corollary 2.5**.**
In systolic complexes, balls around simplices are contractible.
Haglund–Światkowski prove the following.
Proposition 2.6** ([SH], Proposition 4.9).**
A full subcomplex of a systolic complex is convex if and only if is geodesically convex in (i.e. if all geodesics in joining vertices of lie in ).
We record:
Corollary 2.7**.**
In systolic complexes balls around simplices are geodesically convex.
We will need a crucial ”projection lemma”. The residue of a simplex in is the union of all simplices in , which contain .
Lemma 2.8** ([JS], Lemma 7.7).**
Let be a convex subcomplex of a systolic complex and let be a simplex in . Then the intersection of the residue of and of the complex is a simplex (in particular it is nonempty).
Definition 2.9**.**
The simplex as in Lemma 2.8 is called the projection of onto .
The following lemma immediately follows from Definition 2.9.
Lemma 2.10**.**
Let be simplices in for some convex and let be their projections onto . Then .
Definition 2.11**.**
For a pair of vertices in a systolic complex we define inductively a sequence of simplices as follows. Take equal to the projection of onto for . The sequence is called the directed geodesic from to (this notion is introduced and studied in [JS]).
We can extend this construction to any pair , where is a convex subcomplex of and is a simplex. Namely, if for some we have then take to be the projection of onto . If intersects both and then take and then proceed as previously. We call the final the projection of onto . Note that this coincides with Definition 2.9. Observe that if then the projection of onto is equal to .
Finally, recall a powerful observation.
Lemma 2.12** ([JS2], Lemma 4.4).**
Every full subcomplex of a systolic complex is aspherical.
3 Definition of the boundary
Let be a systolic complex. In this section we give two equivalent definitions of the boundary of as a set. We use the notion of Euclidean geodesics which will be introduced in Section 9, but actually we need only its features given by Theorem B and Theorem C. Thus, it is enough to read Sections 1–2 to follow the first part of the article (Sections 3–6). Let be a natural number, which is a universal constant satisfying assertions of both Theorem B and Theorem C.
Remark 3.1**.**
Let be a Euclidean geodesic and let be a vertex in for some . Then there exists a 1–skeleton geodesic such that for . This follows from the fact that , which we use for , and from , which we use for (see Section 1 or Lemma 9.15(i)).
Definition 3.2**.**
Let be vertices of a systolic complex . Let be a geodesic in the –skeleton of between and or let be a –skeleton geodesic ray starting at (then we set ). For , by we denote the Euclidean geodesic between and . We say that is a good geodesic between and or that is a good geodesic ray starting at if for every and every we have (the constant is defined in the beginning of this section).
By we denote the set of all good geodesic rays in . For a given vertex of , by we denote the set of all good geodesic rays starting at .
The following two results are immediate corollaries of Theorem B and Theorem C.
Corollary 3.3**.**
For every two vertices there exists a good geodesic between them.
Proof. Let be the Euclidean geodesic between and . By Remark 3.1, there exists a –skeleton geodesic with . We claim that is a good geodesic. To justify the claim let . Let be the Euclidean geodesic between and . By Theorem B, for every , we have
[TABLE]
which justifies the claim.
Corollary 3.4**.**
Let be good geodesics in . Then for all we have , where .
Proof. Let be the Euclidean geodesics between and, respectively, , . Fix . Pick vertices and which realize the distance to , respectively. Find 1–skeleton geodesics and such that and . Their existence is guaranteed by Remark 3.1. By Theorem C, we have
[TABLE]
as desired.
The following simple corollary of Corollary 3.4 will be useful.
Corollary 3.5**.**
Let be good geodesics in . Then for all we have .
Proof. W.l.o.g. we can assume that . Observe that . Hence, by Corollary 3.4, we have
[TABLE]
Below we define the central object of the article.
Definition 3.6**.**
The (ideal) boundary of a systolic complex is the set of equivalence classes of good geodesic rays, where rays are identified if is bounded above by a constant independent of (one can check this happens exactly when the Hausdorff distance between and is finite). For a good geodesic ray , we denote its equivalence class in by .
In order to introduce topology on we give another definition of the boundary. The two definitions will turn out to be equivalent in the case of a systolic complex with a geometric group action.
Definition 3.7**.**
Let be a vertex of a systolic complex . Then the (ideal) boundary of with respect to the basepoint vertex is the set of equivalence classes of good geodesic rays starting at , where rays are identified if is bounded above by a constant independent of (again this happens exactly when the Hausdorff distance between and is finite). For , we denote its equivalence class in by (we hope this ambiguity of the notation will not cause confusion).
Lemma 3.8**.**
Let . Then iff for all .
Proof. We show that if for some we have , then . Let be as above and be a natural number. Then, by Corollary 3.4, we have
[TABLE]
Since can be chosen arbitrarily large, we get .
In the remaining part of this section we prove equivalence of the above two notions of boundaries in the case of locally finite complexes. Assume that is a locally finite systolic complex. Let be a fixed vertex and let be a good geodesic ray in . For every we choose a good geodesic , guaranteed by Corollary 3.3. Since is finite, for some vertex there are infinitely many such that and . Similarly, since all balls are finite, we obtain inductively vertices satisfying the following. For each there are infinitely many such that and for all we have . For each denote some such by . The following easy facts hold.
Lemma 3.9**.**
The sequence obtained as above is a good geodesic ray. Moreover, for every we have .
Proof. The first assertion follows from the fact that for every the sequence is a subsequence of the good geodesic and hence, by Definition 3.2, it is a good geodesic.
Now we prove the second assertion. Let . Consider the case (the case can be examined analogically). Then for we have . Thus we can apply Corollary 3.4 with to good geodesics and , which yields the following.
[TABLE]
Hence
[TABLE]
where the last inequality follows from , which is the triangle inequality for and .
Corollary 3.10**.**
Let be a locally finite systolic complex and its vertices. Then the map given by is well defined. It is a bijection between and . Its restriction is a bijection between and .
4 Topology on
Let be a systolic complex and be its vertex. In this section we define the topology on the set , which extends the usual topology on the simplicial complex . The idea is to define the topology through neighborhoods (not necessarily open) of points in . The only problem is to define the neighborhoods of points in the boundary.
For a –skeleton geodesic or a geodesic ray , we denote by the combinatorial ball of radius around the subcomplex . Let and be the constants from the previous section.
Definition 4.1**.**
Let be a good geodesic ray in and let (i.e. ) and be natural numbers (in fact we could also allow , but this would complicate some computations later on). By we denote the set of all good geodesics with and good geodesic rays , such that .
By we denote the set .
A standard neighborhood of is the set
[TABLE]
We write , and instead of , and if it is clear what is the basepoint .
Before we define the topology, we need the following useful lemmas. The first one is an immediate consequence of the above definition.
Lemma 4.2**.**
Let and let be natural numbers such that . If then .
Lemma 4.3**.**
Let be a standard neighborhood, let be such that and let be a natural number. Then, for , we have .
Proof. Denote and .
By Lemma 4.2, it is enough to show that for every we have .
Let . By Corollary 3.4 and Lemma 3.8, we have
[TABLE]
Thus and the lemma follows.
The following defines topology on .
Proposition 4.4**.**
Let be the family of subsets of satisfying the following. is open in and for every there is some such that and there is a standard neighborhood . Then is a topology on .
Proof. The only thing we have to check is the following. If and then there is a standard neighborhood of contained in .
Since , for , there are standard neighborhoods such that . Thus, by Lemma 4.3, for any natural there exists , , such that .
Remark 4.5**.**
It is easy to verify that when is –hyperbolic (in the sense of Gromov) then our boundary (with topology induced from ) is homeomorphic in a natural way with the Gromov boundary of .
We still did not prove that the topology defined in Proposition 4.4 is non–trivial. This will follow from the next two lemmas, in which we characterize the intersections with the boundary of the interiors of standard neighborhoods. In particular, we show that is contained in the interior of .
Lemma 4.6**.**
For a set , the intersection consists of those points for which there exists a representative with a standard neighborhood .
Proof. Let be the set of those points for which there exists a representative of with a standard neighborhood .
It is clear that , since is open in the topology defined in Proposition 4.4. We want now to prove the converse inclusion . It is clear that . Thus to prove the lemma we only have to show that is open in (in the topology induced from ).
Let and let its representative be such that the standard neighborhood is contained in . By Lemma 4.3, we can assume that . Choose natural number . We claim that (i.e. that equivalence classes of elements in lie in ). This implies that is open in .
To justify the claim let . To prove that it is enough to establish , since the latter is contained in . By Lemma 4.2, it is enough to show that for every , we have . Let . Denote .
By Corollary 3.4, we have
[TABLE]
Thus and it follows that , which justifies the claim.
Lemma 4.7**.**
Let be a standard neighborhood. Let be such that , where . Then is contained in the interior of .
Proof. By Lemma 4.6, it is enough to show that there exists a standard neighborhood of contained in . Let . By Lemma 4.2, it is enough to show that for , we have . By Corollary 3.4, we have
[TABLE]
as desired.
Below we give a sufficient condition for two standard neighborhoods to be disjoint.
Lemma 4.8**.**
Let and be two standard neighborhoods, with and . If , then .
Proof. By contradiction. Assume .
Case 1: Let . Then, by Definition 4.1, there exist and such that belongs to the interior of both some simplex with vertex and some simplex with vertex , for some . Then these simplices coincide and . By Corollary 3.5, we have
[TABLE]
contradiction.
Case 2: Let and be such that . Then, by Lemma 3.8, we get
[TABLE]
contradiction.
5 Compactness and finite dimensionality
Let be a locally finite systolic complex and let be its vertex. In this section we show that is compact metrizable and (if satisfies some additional local finiteness conditions) finitely dimensional. We also prove that, for a different vertex of , the compactifications and are homeomorphic.
Proposition 5.1**.**
If is locally finite then the space is second countable and regular.
Proof. It is clear that is second countable. We show that is regular.
First we show that is Hausdorff. We consider only the case of two points of the boundary—the other cases are obvious. Let be two boundary points with and . Fix a natural number (for example ). We can find such that . Then, by Lemma 4.7, we have and and, by Lemma 4.8, we get . Thus we get disjoint non–empty open neighborhoods of and .
To show that is regular it is now enough, for every point and every closed subset which does not contain , to find disjoint open sets such that and . Let be as above. The case is obvious thus we consider only the case of , for . Fix some natural . Since is open, by definition of the topology (Proposition 4.4) and by Lemma 4.3, we can find a natural number such that , where . Let and let . Observe that, by Lemma 4.7, we have . Now we define . For each , choose an open set for some vertex in such that . Then we set . By Lemma 4.7, we have , hence . Thus to prove that and are as desired we only need to show that .
First we prove that , for . Let . Then, by Corollary 3.4 and by , we have
[TABLE]
Thus, by Lemma 4.8, .
Now we show that , for . By contradiction, assume . Since , there exist a vertex of the simplex containing in its interior and a good geodesic , where . Then, by Corollary 3.4, we have
[TABLE]
On the other hand, since , there is a vertex such that . Then . Let be a good geodesic. We have , hence by Corollary 3.4 and Corollary 3.5, we get
[TABLE]
Summarizing, we have . It follows that and hence —contradiction.
Corollary 5.2**.**
If is locally finite then the space is metrizable.
Proof. This follows from the Urysohn Metrization Theorem—cf. [Du, Corollary 9.2].
Proposition 5.3**.**
If is locally finite then the space is compact.
Proof. By Corollary 5.2, it is enough to show that every infinite sequence of points in contains a convergent subsequence. Let be a given sequence of points in . If for some there is only finitely many outside the ball (which is finite), then we can find a convergent subsequence. From now on we assume there is no as above.
For every we choose a good geodesic or a good geodesic ray the following way. If then is a good geodesic between and a vertex lying in a common simplex with the point . If then we set for an arbitrary such that and we set . By our assumptions on , for every there exists an arbitrarily large such that . Since is finite, for some vertex there are infinitely many such that and . Let be some such . Similarly, since all spheres are finite, we obtain inductively vertices and numbers satisfying the following. For each there are infinitely many such that and for all we have ; we denote some such by .
Observe that for every the sequence is a subsequence of the good geodesic or the good geodesic ray and hence, by Definition 3.2, it is a good geodesic. Thus every subsequence of the infinite sequence is a good geodesic and again, by Definition 3.2, is a good geodesic ray.
We claim that the sequence of points of converges to , where . To prove the claim it is enough to show (since every open set containing contains some , by Lemma 4.3) that we have , for every . This follows from the equality , which holds for every .
Observe that by the above proof we get the following.
Corollary 5.4**.**
If a locally finite systolic complex is unbounded then its boundary is non–empty.
Below we prove that the bijection defined in Corollary 3.10 extends to a homeomorphism of compactifications coming from different basepoints.
Lemma 5.5**.**
Let be a locally finite systolic complex and let be its vertices. Then the map defined as an extension by identity on of the map is a homeomorphism.
Proof. By compactness (Proposition 5.3) and by Corollary 3.10, we only have to show that is continuous. It is enough to check the continuity at the boundary points. Let be obtained from a good geodesic ray as in the definition of the map . We show that is continuous at . Let , let be a natural number and and let be an open neighborhood of in . We have to show that there exists an open neighborhood of in such that . By Lemma 4.3, there exists such that . Let . By Lemma 4.7, . We claim that —this will finish the proof.
First we show that for we have . For such an choose, by definition of , a good geodesic such that belongs to the interior of a simplex with vertex , where . Let be a good geodesic guaranteed by Corollary 3.3. Then , hence and is defined. By Lemma 3.9 and Corollary 3.4, we have
[TABLE]
This inequality implies that and hence .
Now we show that for we have . Let . Let be obtained from as in the definition of . Then, by Lemma 3.9 and Corollary 3.4, we can perform the same computation as in the previous case to get . Thus and we have completed the proof of and of the whole lemma.
Now we proceed to the question of finite dimensionality of . Let us remind that a simplicial complex is uniformly locally finite if there exists a natural number such that every vertex belongs to at most different simplices. This happens for example when some group acts geometrically on .
Proposition 5.6**.**
Let be a uniformly locally finite systolic complex. Then is finitely dimensional.
Proof. Recall that a space has dimension at most if, for every open cover of , there exists an open cover ( is a refinement of , i.e. every element of is contained in some element of ) such that every point in belongs to at most elements of (i.e. the multiplicity of is at most ).
It is clear that is finitely dimensional. It is thus enough to show that there exists a constant such that for every open (in ) cover of there exists an open cover of of multiplicity at most .
Let be a natural number. Then, by uniform local finiteness, there is a constant such that every ball of radius at most contains at most vertices.
Let be an open cover of in . We construct an open cover of in consisting of interiors of standard neighborhoods such that the multiplicity of is at most .
Let . By the definition of topology (Proposition 4.4) and by Lemma 4.3, for every there exists a standard neighborhood contained in some element of . By Lemma 4.7 we have . By compactness (Proposition 5.3), among such neighborhoods we can find a finite family such that the family of smaller standard neighborhoods covers . Let . Let denote the set of vertices in for which there exists a good geodesic ray starting at and passing through . For each , pick some such good geodesic ray . We claim that the family is as desired.
First we show that covers . Let be an arbitrary good geodesic ray. Then and thus, by Lemma 4.7, .
Now we show that . To prove this it is enough to show that for every there exists such that . Let . Choose such that . By Lemma 4.2, to show that it is enough to show that, for every , we have . Let and denote . By Lemma 3.8, we have . Then, by Corollary 3.4, we have
[TABLE]
Thus and it follows that .
Finally, we claim that the multiplicity of is at most . By Lemma 4.8, if then . Thus multiplicity of is at most the number of vertices in a ball of radius in , i.e. it is at most .
6 The main result
The aim of this section is to prove the main result of the paper—Theorem A (Theorem 6.3).
The following result will be crucial.
Proposition 6.1** ([BeMe, Proposition 2.1], [Be, Lemma 1.3]).**
Let be a pair of finite–dimensional compact metrizable spaces with nowhere dense in , and such that is contractible and locally contractible and the following condition holds:
- •
For every and every open neighborhood of in , there exists an open neighborhood of contained in such that is null–homotopic.
Then is an ER and is a –set in .
Before proving Theorem A we need an important preparatory lemma.
Lemma 6.2**.**
Let and let be a standard neighborhood of in . Then there exists such that and the inclusion map is null–homotopic.
Proof. Let . By Lemma 4.3, there exists such that , so that it is enough to prove the following. For natural there exists such that and the inclusion map is null–homotopic.
Before we start, let us give a rough idea of the proof. Let us restrict to the problem of contracting loops from in (this turns out to be the most complicated case). Let be such a loop. We connect each vertex of by a good geodesic with , and we are interested in the vertex of this geodesic at certain distance from , where . All vertices constructed in this way lie in a certain ball (see Condition 1 below), which is in turn contained in (see Condition 3 below). If we connect these vertices by –skeleton geodesics in the right order we obtain a loop , which lies in the ball considered (Corollary 2.7) and is contractible inside this ball (Corollary 2.5). So we need to find a free homotopy between and , which we construct via intermediate loops . To find that two such consecutive loops are homotopic in , we need Condition 2. This condition guarantees that all relatively small loops by which consecutive differ can be contracted inside .
Let and . We will show that is as desired. Denote . The choice of and guarantees that the following three conditions hold.
Condition 1. Let be a good geodesic with and . Then .
Indeed, let be such that (guaranteed by definition of ). Since , we have, by Corollary 3.5, that
[TABLE]
Thus, by Corollary 3.4, we have
[TABLE]
Condition 2. Let be as in Condition 1. Then, for every we have .
To show this observe that, as in the proof of the previous condition, we have . Now, let be a vertex of . Choose a good geodesic (guaranteed by Corollary 3.3). Since , we have that and is defined. Thus, by Corollary 3.4 and Corollary 3.5, we have
[TABLE]
Thus and it follows that .
Condition 3. We have .
This follows immediately from Condition 2, but we want to record it separately.
The goal. First observe that by Lemma 4.3 and the definition of . Now we show that the map induced by inclusion is trivial, for every . Let be the smallest full subcomplex of containing . Observe that the vertices of lie in . By Condition 2, is contained in . Thus it is enough to show that the map induced by the inclusion is trivial and we may restrict ourselves only to simplicial spherical cycles.
Case (). Let be two vertices of . We will construct a simplicial path in connecting and .
Choose (using Corollary 3.3) good geodesics , . By Condition 2, is contained in and, by Condition 1, we have . Choose a –skeleton geodesic . Since balls are geodesically convex (Corollary 2.7), this geodesic is contained in and hence, by Condition 3, it is contained in .
Then the –skeleton path connects and and is contained in . Therefore the map is trivial.
Case (). Let be a –skeleton loop in . We show that this loop can be contracted within .
Choose good geodesics (guaranteed by Corollary 3.3), for . By , for , we denote . Let . Observe that, by Corollary 3.5, we have (we consider modulo ), for every (we are not interested in smaller ). For these let be arbitrary –skeleton geodesics. Record that .
Thus, for every and for every , we have a –skeleton loop
[TABLE]
of length at most
[TABLE]
Hence . Since balls are contractible (Corollary 2.5), is contractible inside , which is, by Condition 2, contained in . Thus, for , the loops
[TABLE]
for consecutive are freely homotopic in .
Observe that . On the other hand , by Condition 1 and by geodesic convexity of balls (Corollary 2.7). Moreover, since balls are contractible (Corollary 2.5), can be contracted inside , which lies in , by Condition 3. Thus is contractible in . It follows that the map is trivial.
Case (). Since is a full subcomplex of a systolic complex it is, by Lemma 2.12, aspherical and thus and the map in question is obviously trivial.
Theorem 6.3** (Theorem A).**
Let a group act geometrically by simplicial automorphisms on a systolic complex . Then , where is a vertex of , is a compactification of satisfying the following:
* is a Euclidean retract (ER),* 2. 2.
* is a –set in ,* 3. 3.
for every compact set , is a null sequence, 4. 4.
the action of on extends to an action, by homeomorphisms, of on .
Proof. (1. and 2.) By Corollary 5.2, Proposition 5.3, and Proposition 5.6, we have that is a finitely dimensional metrizable compact space.
Since is a simplicial complex, it is locally contractible and, by Theorem 2.3, it is contractible since it is a systolic complex. By the definition of the topology on (c.f. Proposition 4.4), it is clear that is nowhere dense in . Thus we are in a position to apply Proposition 6.1. Let and let be its open neighborhood in .
By definition of the topology (Proposition 4.4) we can find a standard neighborhood , where . By Lemma 6.2, there exists a standard neighborhood (with , by Lemma 4.7) such that the map is null–homotopic. Thus is an ER and is a –set in .
(3.) Let be an open cover of and let be a compact set. We will show that all but finitely many translates , for , are –small.
Let be such that , for some vertex . As in the proof of Proposition 5.6, we can find a natural number , a finite set of vertices and a collection of good geodesic rays with passing through such that the following holds. The family covers and the family is a refinement of . Thus we can find an open cover of such that every is contained in . By compactness—Proposition 5.3—there is a finite subfamily of covering . It follows that there exists a natural number such that . By properness of the action there exists a cofinite subset such that , for .
We claim that, for every , we have , for some . Assertion (3.) follows then from the claim. Let . Since , there exists such that . We show that . Let and let be a good geodesic (which exists by Corollary 3.3) such that is a vertex of the simplex containing in its interior. Since there exists a good geodesic , such that . We have and . Hence, by Corollary 3.5, we have
[TABLE]
Thus and hence . It follows that . Since was arbitrary we have that elements of are –small and thus they are –small.
(4.) For we define a map as follows. For let and for let . This is obviously a well defined homeomorphism.
We extend the action of on to by the formula , for . By Lemma 5.5, the map is a homeomorphism. To see that , for , pick some representative of . We need to show that
[TABLE]
Recall that, by Lemma 3.9, mappings and displace representative rays by a finite Hausdorff distance. Hence is the class of rays starting at at a finite Hausdorff distance from . On the other hand, is the class of rays starting at at a finite Hausdorff distance from , hence as well as is the class of rays (starting at, respectively, and ) at a finite Hausdorff distance from . This proves the desired equality.
Hence we get an extension of the action of on to an action on by homeomorphisms.
7 Flat surfaces
With this section we start the second part of the article, in which we define Euclidean geodesics, establish Theorem B and Theorem C. Before we define Euclidean geodesics, we first need to study, as mentioned in Section 1, the minimal surface spanned on a pair of directed geodesics connecting given vertices. The tools for this are minimal surfaces (Section 7) and layers (Section 8).
In this section we recall some definitions and facts concerning flat minimal surfaces in systolic complexes proved by Elsner [E],[E3].
Definition 7.1**.**
The flat systolic plane is a systolic 2–complex obtained by equilaterally triangulating Euclidean plane. We denote it by . A systolic disc is a systolic triangulation of a 2–disc and a flat disc is any systolic disc , which can be embedded into , such that is embedded isometrically into 1–skeleton of . A systolic disc is called wide if is a full subcomplex of . For any vertex the defect at (denoted by def) is for , and for , where is the number of triangles in containing . It is clear that internal vertices of a systolic disc have nonpositive defects.
We will need the following easy and well known fact.
Lemma 7.2** (Gauss-Bonnet Lemma).**
If is any triangulation of a 2–disc, then
[TABLE]
Flat systolic discs can be characterized as follows.
Lemma 7.3** ([E], Lemma 2.5).**
A systolic disc is flat if and only if it satisfies the following three conditions:
(i) has no internal vertices of defect
(ii) has no boundary vertices of defect
(iii) any segment in connecting vertices with defect contains a vertex of defect .
Now we recall another handful of definitions.
Definition 7.4**.**
Let be a systolic complex. Any simplicial map , where is a triangulation of a 2–disc, is called a surface. We say that is spanned on a loop , if . A loop is triangulable, if there exists a surface spanned on , such that all the vertices of are in . A surface is systolic, flat or wide if the disc satisfies the corresponding property. If is injective on and minimal (the smallest number of triangles in ) among surfaces with the given image of , then is called minimal. A geodesic in is called neat if it stays out of except possibly at its endpoints. A surface is called almost geodesic if it maps neat geodesics in isometrically into .
The following is part of the main theorem of [E].
Theorem 7.5** ([E], Theorem 3.1).**
Let be a systolic complex. If is a wide flat minimal surface in then is almost geodesic.
We will also use the following handy fact, whose proof can be extracted from [E3]. In case has length 2 it follows immediately from 6–largeness.
Proposition 7.6** ([E3], Proposition 3.10).**
Let be a systolic complex and a wide flat minimal surface. Let be a neat 1–skeleton geodesic in , which is contained in a straight line. Then for any 1–skeleton geodesic in with the same endpoints as there is another minimal surface such that and on the vertices of outside .
8 Layers
In this section we introduce and study the notion of layers for a pair of convex subcomplexes of a systolic complex. If those subcomplexes are vertices , then the layer is the span of all vertices, in 1–skeleton geodesics , at distance from (c.f. Definition 8.1). In particular, simplices of the directed geodesics between and (c.f. Definition 2.11), as well as the simplices of Euclidean geodesics (which we construct in Section 9) lie in appropriate layers.
On the other hand, layers in systolic complexes seem to be interesting on their own.
Definition 8.1**.**
Let be convex subcomplexes of a systolic complex and . For we define the layer between and as the subcomplex of equal to . We will denote it by (or shortly when are understood).
Remark 8.2**.**
are convex, since they are intersections of convex (see remarks after Definition 2.4).
Lemma 8.3**.**
(i) , for .
(ii) , for . In particular , for .
Proof. (i) W.l.o.g we only need to prove that . Take a vertex . Then we have and , while . Thus by the triangle inequality we have , as desired.
(ii) By (i) we have that , thus we only need to prove that . Let be a vertex in . Since, by (i), we have , there is a vertex at distance from . Since , we have . Thus and .
Now we study the properties of layers.
Lemma 8.4**.**
For we have that is –large.
Proof. Suppose the layer is not –large. Then there exists an embedded cycle in (denote its consecutive vertices by ) which is a full subcomplex of .
Denote . We have that . Notice that is a full subcomplex of , because there are no edges in between vertices in and vertices in .
Observe that is contractible in (and similarly in ). Indeed, by Lemma 8.3(i) we have that . Thus we can project the edges of onto (c.f. Definition 2.9). If we choose a vertex in each of these projections, we get, by Lemma 2.8, that these vertices form a loop. This loop is homotopic to in . Moreover, since is contractible (by remarks after Definition 2.4) it follows that is contractible in (and similarly in ), as desired. The simplicial sphere formed of these two contractions is contractible in as full subcomplexes of are aspherical (Lemma 2.12).
Now use Meyer-Vietoris sequence of the pair . Since is the image of under it follows that the cycle is homological to zero in itself. This is a contradiction.
Lemma 8.5**.**
Let be maximal simplices in the layer for some and . Then or or .
Proof. W.l.o.g. assume that . Suppose the lemma is false. Then there exist vertices . By Lemma 8.3(ii) we have that . Denote by some vertices in the projections (c.f. Definition 2.9) of onto . We have , because both and are neighbors of and the projection of (c.f. Lemma 8.3(ii)) onto is a simplex (Lemma 2.8). Now we will argue that we can assume that is an edge. In case consider the 4–cycle . It must have a diagonal. We can then assume w.l.o.g. that is an edge. In case we also have that is an edge. In both cases it follows that belongs to the simplex which is the projection of (c.f. Lemma 8.3(ii)) onto . This simplex also contains . But , which contradicts the maximality of .
Corollary 8.6**.**
Let be the following simplicial complex: the trapezoid build of three triangles . Then there is no isometric embedding of into , for .
Proof. Extend the images of those three triangles to maximal simplices and apply Lemma 8.5.
Corollary 8.7**.**
Let . Let for vertices such that and . Then, for any 1–skeleton geodesics connecting with and with , respectively, and for any such that , we have that (i.e. is an edge or ).
Proof. We will prove the corollary by induction on . First observe that since is –large (Lemma 8.4), the loop is triangulable and there exists a diagonal cutting off a triangle. There are only four possibilities for this diagonal and we can w.l.o.g suppose this diagonal is . Now since and both and lie in the projection of onto , then by Lemma 2.8 either is an edge or .
Now we start the induction. If and the loop is embedded, then it has a diagonal. The rest of the required inequalities follows from applying twice Corollary 8.6.
Suppose that for the corollary is already proved. Then applying it to the pair yields all the required inequalities except for the estimate on . But this follows from Corollary 8.6 applied to the trapezoid .
Corollary 8.8**.**
If are edges in , for some , such that and , then .
Proof. By contradiction.
Case . If (if there is a diagonal in the square ) then Corollary 8.7 applied to in place of , gives , contradiction.
Case . Again apply Corollary 8.7, with , getting , contradiction.
Below we present another important property of layers. Since it will not be needed in the article, we do not include the proof. Denote for some .
Lemma 8.9**.**
* is –large.*
We end with a simple, but useful observation.
Lemma 8.10**.**
For any edges such that , where , we have that .
Proof. By contradiction. Suppose, w.l.o.g., that . Hence lies on a 1–skeleton geodesic . Thus, by convexity of layers (Remark 8.2) and by Proposition 2.6, we have that lies in , which is, by Lemma 8.3, disjoint with , contradiction.
9 Euclidean geodesics
In this section we define, for a pair of simplices as below, a sequence of simplices in the layers between and , which can be considered as a ”Euclidean” geodesic joining and . Unlike the directed geodesics defined by Januszkiewicz and Światkowski (see Definition 2.11), Euclidean geodesics are symmetric with respect to and .
The definition requires a lenghty preparation. Roughly speaking, we start by spanning a minimal surface between directed geodesics from to and from to . We observe that this surface is flat whenever the two directed geodesics are far apart (we call the corresponding layers thick). Next we show that this ”piecewise” flat surface is in some sense unique. This occupies the first part of the section, up to Definition 9.9. Then we look at the geodesics in the Euclidean metric in the flat pieces and use them to define Euclidean geodesics in systolic complexes, c.f. Definition 9.12. Finally, we establish some of their basic properties.
The setting, which we fix for Sections 9—13 is the following. Let be simplices of a systolic complex , such that for some natural we have . Let and be sequences of simplices in , such that for we have that span a simplex and span a simplex. In particular, lie in the layer between and (c.f. Definition 8.1).
Note that if is the directed geodesic from to and is the directed geodesic from to (c.f. Definition 2.11), then the above condition is satisfied. This special choice of will be very important later and we will frequently distinguish it.
Definition 9.1**.**
For the thickness of the layer for is the maximal distance between vertices in and in . If the layer for has thickness we say that the layer for is thin, otherwise we say that the layer for is thick. If are directed geodesics from to and from to , respectively, then we skip ”for ” for simplicity.
Caution. Perhaps, to avoid confusion, we should not have used the word ”layer” in the above definition, since we are in fact only checking the position of w.r.t. . Even if the layer between and is large, it can happen that the thickness of the layer for is small. However, we decided that this terminology suits well our approach, in which we will be mostly interested in the part of the layer between and , which lies between and .
Definition 9.2**.**
A pair , where is called a thick interval (for ) if the layers and (for ) are thin, , and for every , such that , the layer (for ) is thick. We say that the thick interval contains if .
Lemma 9.3**.**
(i) The thickness of consecutive layers varies at most by 1.
(ii) If is a thick interval (for ), then are disjoint.
Proof. Both parts follow immediately from Lemma 8.10.
Definition 9.4**.**
Let be a thick interval (for ). Let vertices be such that for each the distance is maximal (i.e. realize the thickness of the layer ). By Lemma 9.3(ii) the sequence is an embedded loop, thus we can consider a minimal surface spanned on this loop (c.f. Definition 7.4). We say that is a characteristic surface (for the thick interval ) and is a characteristic disc.
Lemma 9.5**.**
For , if distances equal the thickness of the layer then also equals the thickness of the layer , i.e. if vertices realize the thickness in some pairs, then they also realize the thickness as a pair.
Proof. Immediate from definition of thickness and Corollary 8.8.
The lemma below summarizes the geometry of characteristic discs, which we need to introduce the concept of a Euclidean geodesic. The special features of characteristic discs, in the case that are directed geodesics, will be given in Lemma 9.16 at the end of this section.
Let be a characteristic surface. Denote by the preimages of in , respectively. This notation will be fixed for the entire article. Let us point out that we use numbers to number the layers in (c.f. Definition 8.1) between and , instead of , for the sake of clarity.
Lemma 9.6**.**
(i) (and thus the characteristic surface ) is wide and flat,
(ii) if we embed , then the edges and are parallel and consecutive layers between them are contained in consecutive straight lines (treated as subcomplexes of ) parallel to the lines containing and .
Proof. (i) To prove wideness it is enough to show that any nonconsecutive vertices of the boundary loop are at distance . Since the layers , where , are thick (for ), the only possibility for this to fail is that (w.l.o.g.) for some . If this happens, then both and lie in the projection of onto the layer between and (the projection is defined by Lemma 8.3(ii)), hence they are neighbors (Lemma 2.8), which contradicts . Thus a characteristic disc is wide.
Before proving flatness, we need the following general observation. If is a 1–skeleton geodesic, which is in the boundary of a triangulation of a disc, then the sum of the defects at the vertices in the interior of is . Moreover, all the defects at these vertices are and each two vertices of positive defect are separated by a vertex of negative defect.
To prove flatness, compute possible defects at the boundary vertices of . By wideness, they are at . Moreover, their sum over the interior vertices of each of the 1–skeleton geodesics is (they are 1–skeleton geodesics, since their images are). Thus Gauss–Bonnet Lemma 7.2 implies that the defects of the interior vertices are equal to zero, the sums of the defects over the vertices equal 1 each and the defects at are equal to .
We now want to say more about the defects at . Up to now we know that their sum is 1, they equal or and each two vertices of positive defect are separated by a vertex of negative defect (since is a 1–skeleton geodesic). This implies that the defects equal alternatingly with possible [math]‘s between them. The same holds for the defects at . Thus, by Lemma 7.3 (characterization of flatness), the characteristic disc is flat, i.e. we have an embedding isometric on the 1–skeleton.
(ii) By the computation of defects in the proof of (i) we get that the edges and are parallel in . We also get that , for are at combinatorial distances from the lines containing the edges . Hence lie on the appropriate line parallel to and the vertices of split into families lying on geodesics . By convexity of layers, Remark 8.2, (or by direct observation) these geodesics are equal to the layers.
When speaking about the layers in between and , we will often skip ”between and ”.
Remark 9.7**.**
Denote the layer in (between and ) by . Then is contained in the layer in between and . This follows from
[TABLE]
The next lemma summarizes some uniqueness properties of characteristic surfaces for a fixed thick interval .
Lemma 9.8**.**
(i) A characteristic surface is almost geodesic. In particular, it is an isometric embedding on the 1–skeleton of a subcomplex spanned by any pair of consecutive layers between and in .
(ii) A characteristic disc does not depend (up to isometry) on the choice of and the choice of a characteristic surface.
If we have two characteristic surfaces , then after identifying the characteristic discs (which is possible by (ii)) we have that
(iii) for any vertices at distance , and are also at distance , i.e. for any two characteristic surfaces we can substitute an image of a vertex of the first surface with the corresponding image in the second and get another characteristic surface,
(iv) for any vertex , and are at distance .
Proof. (i) This follows from Elsner‘s Theorem 7.5, since, by Lemma 9.6(i), a characteristic disc is flat and wide. The second part follows from the fact that any two vertices in a same or consecutive layers in can be connected by a neat geodesic, which can be verified by direct observation.
(ii) Observe that, by Lemma 9.6(ii), the isometry class of is determined by the distances , , for , which are equal, by (i), to , respectively. The value equals the thickness of the layer , so it does not depend on the choices. To prove the same for , consider two characteristic surfaces constructed for choices , where . We will prove that . We restrict ourselves to proving the first equality (the second is proved analogically). By Lemma 9.5 we have that is the thickness of the layer . Thus there is a characteristic surface spanned on a loop passing through . Hence, by (i), the distance is determined by and , thus it is the same as , as desired.
(iii) If and are both boundary vertices, then this is obvious. Otherwise, w.l.o.g. assume that is an interior vertex of . Suppose lies in the layer (we denote it by ) in between and . Denote the thickness of the layer for by .
First consider the case that . By Remark 9.7 we have that and lie in the layer in between and . By Lemma 9.5 we have that . Hence Corollary 8.7 applied to and gives , as desired.
Now, w.l.o.g., consider the remaining case that is in the layer (denoted by ) in between and . Denote by the common neighbors of in , respectively, and by the neighbor of in different from . Then, from the previous case, we have that is a loop of length 5, hence it is triangulable. By (i), all equal 2, hence we obtain , as desired.
Observe that this proof actually implies Proposition 7.6 in the case that for some .
(iv) For boundary vertices this is obvious. For an interior vertex , let be its neighbors in a common layer in between . Then, by (iii), we have that is a loop of length 4. Moreover, by (i), we have that . Thus , as desired.
As a corollary, the following definition is allowed.
Definition 9.9**.**
Let be a simplex of the characteristic disc for some thick interval (for ). Its characteristic image is a simplex in , denoted by , which is the span of the images of under all possible characteristic surfaces. Note that is a simplex by Lemma 9.8(iii,iv), and if , then , i.e. respects inlucions. The characteristic image of a subcomplex of is the union of the characteristic images of all its simplices. We call this assignment the characteristic mapping.
If is a vertex in , we denote by the vertex such that . We claim that this vertex is unique. Indeed, characteristic images of different layers in between are disjoint since, by Remark 9.7, they lie in different layers in between , disjoint by Lemma 8.3. Moreover, by Lemma 9.8(i,iii), we have that for any characteristic surfaces and any vertices in a common layer in . This justifies the claim. If is a simplex in , we denote by the span of the union of over all . We have that is a simplex, by Remark 9.7, Lemma 8.3, and Lemma 9.8(i,iii). If is a subcomplex of , we denote by the union of over all .
Having established the uniqueness properties of characteristic surfaces, we start to exploit the structure of the corresponding characteristic discs. From now on, up to the end of Section 13, unless stated otherwise, assume that are the directed geodesics between .
Definition 9.10**.**
Let be a thick interval and let be its characteristic disc. We will define a sequence of simplices , where , which will be called the Euclidean diagonal of the characteristic disc .
Let be points (barycenters of edges) on the straight line segments at distance from , respectively. In particular . Consider the closed polygonal domain enclosed by the piecewise linear loop with consecutive vertices . Note that, since is simply–connected, it is with the Euclidean path metric induced from identified with . We call a modified characteristic disc. Let be the geodesic joining to in . We call a CAT(0) diagonal of . For each , among the vertices of lying in the interior of the 1–skeleton geodesic find the ones nearest to . For each this is either a single vertex or two vertices spanning an edge (if goes through its barycenter and are not some of its vertices). We put equal to this vertex or this edge, accordingly.
At first sight it might seem strange that in the above definition we pass to and take the geodesic there instead of doing it in itself. However, this construction allows us to exclude from being in , which a careful reader will find to be a necessary condition for the arguments of the combinatorial Proposition 10.2 to be valid.
Here are some basic properties of the Euclidean diagonals.
Lemma 9.11**.**
(i) Each pair of consecutive , for , spans a simplex.
(ii) span a simplex and span a simplex.
Proof. Part (ii) is obvious, since we excluded from being in . To prove (i), consider oriented in such a way that are horizontal, this is possible by Lemma 9.6(ii). Moreover, Lemma 9.6(ii) yields that the boundary of consists of line segments at angle from the vertical direction. Let be as in Definition 9.10. It is a broken line with vertices at the boundary of .
We claim that any line segment of is at angle from the vertical direction. First we prove that this angle is . Otherwise, let be an endpoint of such a line segment. Obviously is different from the endpoints of . The interior angle at between the segment of and any of the boundary line segments of is , which contradicts the fact that is an interior vertex of a geodesic . Thus we proved that any line segment of is at angle from the vertical direction.
If for some line segment of this angle equals , then by the previous considerations the whole is in fact a straight line at angle from the vertical. This implies that the defects at all vertices in or all vertices in are zero. Contradiction.
Now part (i) follows from the following observation, whose proof is easy and is left for the reader. Consider two consecutive horizontal lines in . Let be some straight line segment joining points at angle from the vertical direction. Then there exist two 2–simplices in such that and . Moreover, it cannot happen simultaneously that and .
Thus we can finally introduce the main definition of this section.
Definition 9.12**.**
We define a sequence of simplices , where , which is called the Euclidean geodesic between , as follows. For each , if the layer is thin, then we take as the span of and .
If the layer is thick, consider the thick interval which contains . Let be an appropriate simplex of the Euclidean diagonal of the characteristic disc for (c.f. Definition 9.10). We take (c.f. Definition 9.9).
Remark 9.13**.**
In the above setting, we have , by Lemma 9.3(ii). Hence .
Remark 9.14**.**
By the symmetry of the construction, the Euclidean geodesic between and becomes the Euclidean geodesic between and if we take the simplices of the sequence in the opposite order.
Here is the justification for using the name ”geodesic” in Definition 9.12.
Lemma 9.15**.**
(i) For any we have that .
(ii) For any if the layer or the layer is thick, then and span a simplex.
(iii) For any such that there exists such that the layer is thick, and for any vertices , we have .
Proof. Assertion (ii) follows from Lemma 9.11(i,ii), Remark 9.13 and Lemma 9.8(iii,iv).
To prove assertion (i), say the first inclusion, observe that for any we have . Hence, assertion (ii) gives already, for any , that . Then follows from Remark 9.7 and Lemma 8.3(ii).
To prove part (iii), assume that (other cases are easier). Take any vertex . Then, by (i), there are vertices such that . By (ii) (and (i)), we have . Hence and by (i) we have , as desired.
Now we state an extra property of characteristic discs in the case that (but not necessarily) is the directed geodesic. This property was not necessary for Definition 9.12, but will become indispensable in the next section.
Lemma 9.16**.**
(i) If the defect at some , where , equals , then the defect at equals 1.
(ii) The defect at equals 1.
Proof. (i) Proof by contradiction. Suppose the defect at some , where , equals , and the defect at equals 0. Denote by the vertex next to on the 1–skeleton geodesic and by the vertex next to on the 1–skeleton geodesic . We aim to prove that, for any characteristic surface , belongs to . Suppose for a moment we have already proved this. Then, since by Lemma 9.8(i) we have and at the same time , we get a contradiction.
Now we prove that . By Remark 9.7, lies in the layer between and . Now by definition of projection (c.f. Definition 2.9) we need to prove that is a neighbor of each . Case is obvious, so suppose . Since, by definition of thickness, , we have by Lemma 8.7 (applied to and to in case of or to in case of ) that . Considering the loop , since (Lemma 9.8(i)), we get , as desired.
(ii) By contradiction. Denote by the vertex between and on the 1–skeleton geodesic . Since (see Remark 9.13), we have by Remark 9.7 and Lemma 9.8(iii) that belongs to . By Lemma 9.8(i) we have . At the same time , contradiction.
We will repeat some steps of this proof later on in the proof of Lemma 10.3. We decided, for clarity, not to interwind these two proofs.
As a consequence of Lemma 9.16, we get the following lemma, whose proof, similar to the proof of Lemma 9.11, we omit. Here we assume that both are directed geodesics.
Lemma 9.17**.**
If then the diagonal in crosses each line orthogonal to the layers transversally.
10 Directed geodesics between simplices of Euclidean geodesics
In this section we start to prove a weak version of Theorem B, which concerns one of the main properties of Euclidean geodesics. Roughly speaking, the theorem says that pieces of Euclidean geodesics are coarsely also Euclidean geodesics.
We keep the notation from the previous section. The simplices are in this section the directed geodesics between .
Theorem 10.1** (weak version of Theorem B).**
Let be simplices of a systolic complex , such that for some natural we have (as required in the definition of the Euclidean geodesic). Let be the Euclidean geodesic between and . Take some and consider the simplices of the Euclidean geodesic between and (we can define it by Lemma 9.15(i)). Then for each we have .
The proof of Theorem 10.1 splits into two steps. The first step is to prove that directed geodesics between and stay close to the union of characteristic images of all characteristic discs (for ). This is the content of Proposition 10.2, whose proof occupies the rest of this section.
The second step is to check that characteristic images for the directed geodesics between and also stay close to the union of characteristic images for . Properties of layers actually imply that characteristic discs of the former are embedded into characteristic discs of the latter, modulo small neighborhood of the boundary. So everything boils down to the fact that Theorem 10.1 is valid for subspaces of the Euclidean plane. We carry out this program in the next section. We also indicate there an argument, how to promote Theorem 10.1 to Theorem B, with a reasonable constant .
A complete alternative proof of Theorem B, with a worse constant , is obtained as a consequence of Proposition 12.1. We present it at the end of Section 12. We advise the reader to have a look at the proof of Theorem 10.1 via Proposition 10.2. This proof is straightforward and allows us to introduce gradually some concepts needed later. However, to save time, one can skip the remaining part of Section 10, go over the definitions in Section 11 and then go directly to Section 12.
For each thick layer contained in a thick interval (for ; from now on we often skip ”for ”), denote by the appropriate simplex (in the corresponding characteristic disc ) of the directed geodesic from , if , or otherwise, to , if , or otherwise. The simplices of the directed geodesic from to satisfy the following.
Proposition 10.2**.**
Let .
(i) If the layer is thin, then contains or is contained in ,
(ii) if the layer is thick, then contains or is contained in .
Before we give the proof of Proposition 10.2, we need to establish some necessary lemmas. The first one describes the position of with respect to the characteristic image. Like in Lemma 9.16, here does not need to be the directed geodesic.
Lemma 10.3**.**
For a thick layer let be the vertex, which is a neighbor of on the 1–skeleton geodesic in the characteristic disc for the thick interval containing . If the defect at equals 1, then . Otherwise .
Proof. First of all follows from the definition of thickness and Proposition 7.6 (one could also verify this by hand, similarly like in the proofs of Lemma 9.6(iii) and Lemma 9.16(i)). Suppose the defect at is . Hence , by Lemma 9.16(i,ii). The inclusion is obvious and the converse inclusion follows from and from Lemma 9.8(i).
Now suppose the defect at equals . If the layer is thick, then the defect at is and we apply what we have just proved to get . If the layer is thin we get immediately that (Remark 9.13). In both cases using Remark 9.7, Lemma 9.8(iii), and the definition of projection we get , as desired.
As a corollary we get the following technical lemma.
Lemma 10.4**.**
Suppose do not satisfy for any thick interval or if they violate this then . Then the projection of onto equals .
Proof. To justify speaking about the projection of onto we must show that . The simplex is outside by Remark 9.7 and Lemma 8.3. Thus we only need to check that .
To verify this, we prove that . If the layer is thin then this follows from Lemma 9.15(i). If the layer is thick, then denote by the thick interval containing . By Lemma 10.3 we have ( as in Lemma 10.3). Thus it is enough to establish the inclusion . If , then this follows from our assumptions. If , then from Remark 9.13 and Lemma 9.15(i) we have
[TABLE]
as desired.
Hence the projection of onto is defined. Denote it by . Since , we have . For the converse inclusion we need , which we have just proved.
The next lemma is valid for any , not necessarily directed geodesics.
Lemma 10.5**.**
Let be an edge in the layer of (between ), such that has three neighboring vertices in the layer . Let be a vertex in the residue of (for some characteristic surface ) in the layer between in . Then , where is the vertex in the layer of in the residue of .
Proof. Denote by the neighbors of in the layer of different from , and let . We claim that are neighbors of . Indeed, let be the vertex in , which is a neighbor of . Let . Observe that both lie in the projection of onto (by Remark 9.7), hence, by Lemma 2.8, they are neighbors, as desired. Analogically, are neighbors. Thus, by the easy case of Proposition 7.6, , as required.
The following lemma describes the behavior of the simplices appearing in the statement of Proposition 10.2. The proof of Lemma 10.6 requires Lemma 9.16(i,ii), apart from this it is straightforward and we skip it. For the same reason we will usually not invoke it in the proof of Proposition 10.2.
Lemma 10.6**.**
Let be a characteristic disc for some thick interval . Suppose for some we have simplices in the layers respectively between in . Suppose that and . Moreover, assume that is an interior vertex of or an edge disjoint with the boundary or . Assume that is an interior vertex or an edge disjoint with boundary or . Let be the directed geodesic in joining to (in particular ). Then:
(i) If is an edge, then is the unique vertex, which is in the residue of in the layer .
(ii) If and the defect at equals [math], then .
(iii) If is a vertex with two neighbors in the layer , both at distance from , then is an edge spanned by these two vertices.
(iv) If is a vertex with two neighbors in the layer , but only one of them at distance from , then is this special vertex.
Moreover, never equals . If is an edge containing then the defect at is . If , then the defect at is not equal to 1, except possibly for the cases .
Now we are ready for the following.
Proof of Proposition 10.2. We will prove by induction on , for , the following statement, which, by Lemma 10.6 and Lemma 10.3, implies the proposition.
Induction hypothesis. (1) If the layer is thick and is an edge disjoint with the boundary or meeting the boundary at a vertex of defect , then is contained in ,
(2) if the layer is thick and is a non–boundary vertex, then contains ,
(3) if the layer is thick and is a boundary vertex or an edge intersecting the boundary at a vertex of defect 1, or the layer is thin, then contains or is contained in .
For the hypothesis is obvious. Suppose it is already proved for some . We would like to prove it for . First suppose that the layer is thick and is an edge disjoint with the boundary or meeting the boundary at a vertex of defect (case (1)). Then is a vertex. If it is a boundary vertex, then . By the induction hypothesis, since the defect at is not 1, , moreover, by Lemma 10.3 we have , hence . Hence, by Lemma 2.10, is contained in the projection of onto , which in this case equals by Lemma 10.4. Thus , as desired.
Now, still assuming that the layer is thick and that is an edge disjoint with the boundary or meeting the boundary at a vertex of defect , suppose that is not a boundary vertex. Let be any vertex in . Our goal is to prove that . By induction hypothesis we know that . Since lies in the layer between , by Lemma 9.7, we can apply Lemma 10.5 with . Hence we get , as desired.
Thus we have completed the induction step in case (1), i.e. for the layer thick and an edge disjoint with the boundary or meeting the boundary at a vertex of defect .
Now suppose that the layer is thick and is a non–boundary vertex (case (2)). Then it has two neighbors in the layer of , suppose first that both of them are at distance from (we put if ). Then is the edge spanned by those two vertices. If it intersects the boundary, the defect at the boundary vertex is not 1. Thus we must show that is contained in . But by induction hypothesis we know that contains . Thus, by Lemma 2.10, it is enough to observe that . This follows from .
If one of the two neighbors of in the layer is not at distance from , then is the second neighbor, it is a non–boundary vertex (unless , which will be considered in a moment) and . Thus we must show that contains . Let be a vertex in . Then lies on a 1–skeleton geodesic of length from some vertex of to some vertex . We claim that if is an edge, then the vertex is the vertex closer to then the other vertex of . Indeed, let be the vertex closer to . Since and this distance is realized by a neat geodesic, hence by Lemma 9.8(i) we have . This proves the claim. Thus we can apply Lemma 7.6 to and , and get , as desired.
Now we come back to the case and a non–boundary vertex. By induction hypothesis we have . By Lemma 9.15(i) we have that (Remark 9.13) lies in . Hence, by Lemma 2.10, we have that contains , as desired.
Thus we have completed the induction step in case (2), i.e. for the layer thick and a non–boundary vertex.
Now consider the case that the layer is thick and is a boundary vertex or the layer is thin, but the layer is thick (in this case put ). In both cases . First consider the case that the defect at is or the layer is thin. If the hypothesis of Lemma 10.4 are not satisfied, then we can finish as in the previous case (no matter what is the direction of the inclusion given by the induction hypothesis) getting . Otherwise, is the edge spanned by two neighbors of in the layer . By Lemma 9.16(i,ii) the defect at equals 1. Hence we want to prove that either contains or is contained in . We know, by the induction hypothesis, that contains or is contained in , hence it is enough to use Lemma 2.10 and Lemma 10.4.
Now assume that either the layer is thick and is a boundary vertex of defect 0 or an edge intersecting the boundary at a vertex of defect 1, or the layer is thin and the layer is also thin. Similarly, as before, we have that that contains or is contained in and we want to prove that contains or is contained in . This follows from Lemma 2.10 and Lemma 10.4.
Thus we have exhausted all the possibilities for case (3) and completed the induction step.
11 Euclidean geodesics between simplices of Euclidean geodesics
In this section we complete the proof of Theorem 10.1. Its first ingredient is Proposition 10.2, proved in section 10. The second ingredient is easy 2–dimensional Euclidean geometry, which we present as a series of lemmas in this section. Throughout the section, we will be treating characteristic discs simultaneously as simplicial complexes and metric spaces.
We start with extending in various ways the notion of a characteristic disc and surface.
Definition 11.1**.**
A generalized characteristic disc for an interval , where , is a closed (i.e. simply connected) subspace of with the following properties. Its boundary is a piecewise linear loop with vertices (possibly ) , such that for the straight line segments (or points) are contained in consecutive parallel lines at distance . We also require, if is oriented so that are horizontal, that lies to the left of , or .
A restriction of a generalized characteristic disc to the interval , where , is the generalized characteristic disc enclosed by the loop . We will denote it by . If a generalized characteristic disc comes from equipping a systolic –complex with the standard piecewise Euclidean metric, then we call it a simplicial generalized characteristic disc.
Remark 11.2**.**
Characteristic discs (resp. modified characteristic discs, c.f. Definition 9.10) with the standard piecewise Euclidean metric are simplicial generalized characteristic discs (resp. generalized characteristic discs).
Definition 11.3**.**
Suppose we have simplices in the layer between (not necessarily the simplices of the directed geodesics) defined only for , where , and for we have that span a simplex and span a simplex. Suppose that for the maximal distance between vertices in and in is . Then we define a * partial characteristic disc* and a partial characteristic surface in the following way.
Extend to all so that for and span simplices for , and . (This is possible, since, by example, we may issue directed geodesics from to and from to .) Obviously, , lie in the layer between for all . Let be the thick interval for extended containing . Let be a characteristic surface for . Then we call a partial characteristic disc (which is a simplicial generalized characteristic disc) and a partial characteristic surface.
Caution. A characteristic surface , where is a characteristic disc for a thick interval for (as in Definition 9.4) is not a partial characteristic surface for . This is because the layers are thin. But if , then already restricted to is a partial characteristic surface.
Next we show that partial characteristic surfaces satisfy most of the properties of characteristic surfaces. Fix an interval and simplices as in Definition 11.3. Let be a partial characteristic surface, as above.
Lemma 11.4**.**
(i) (and thus ) is flat,
(ii) if we embed , then and are parallel and the consecutive layers between them are contained in consecutive straight lines parallel to and .
(iii) is an isometric embedding on 1–skeleton of a subcomplex spanned by any pair of consecutive layers between and in .
(iv) does not depend on the choice of for and , the choice of for , and the choice of .
If we have two partial characteristic surfaces , then after identifying partial characteristic discs (which is possible by (ii)) we have that
(v) for any vertices at distance , and are also at distance ,
(vi) for any vertex , and are at distance .
(vii) lies in the layer between and .
Proof. Assertions (i) and (ii) follow immediately from Lemma 9.6(i,ii). Assertion (iii) follows from Lemma 9.8(i). To prove (iv) notice that is determined by the distances for and for , by (iii). Hence, if we fix and for , then does not depend on the extension of . On the other hand, if we fix such an extension, then do not depend on the choice of , by Lemma 9.8(ii).
It is a bit awkward to try to obtain assertion (v) as a consequence of Lemma 9.8(iii). Let us say, instead, that assertion (v) follows immediately from the proof of Lemma 9.8(iii). Similarly, assertion (vi) follows from the proof of Lemma 9.8(iv). Assertion (vii) follows directly from Remark 9.7.
Definition 11.5**.**
We define the partial characteristic image of a simplex in the partial characteristic disc as the span of over all partial characteristic surfaces . By Lemma 11.4(v,vi), is a simplex. We call this assignment the partial characteristic mapping. Like in Definition 9.9 we can consider also the assignment .
Definition 11.6**.**
Let be a generalized characteristic disc and , be two paths connecting some points on to points on such that intersections of , with are unique for each . We say that are –close if they intersect in points at distance for each .
The following lemma describes the possible displacements of geodesics in characteristic discs when perturbing the boundary and the endpoints.
Lemma 11.7**.**
Let be two generalized characteristic discs for such that for each we have (and the order is ) and . Then for any points such that , the geodesics from to in and from to in are –close in .
Proof. Denote by the geodesics from to in and from to in respectively. Denote by the set of points in at distance from in the direction parallel to (i.e. the intersection with of the union of translates of by a distance in the direction parallel to ), and by the intersection .
Observe that is connected, since for each the set is nonempty and the intersection of with each of the parallelograms is an intersection of two parallelograms, hence convex and connected. We claim that is convex in . To establish this, we need to study the interior angle at vertices of outside . The only possibility for angle is at the horizontal translates of break points of . But since is a geodesic, then each of its break points lies on the boundary of , and the translate, for which possibly the angle is , lies outside . Thus the claim follows. Hence (by connectedness) is convex in . Thus and we are done.
Let us prepare the setting for the next lemma. It will help us deal with the data given by Proposition 10.2, which is, roughly speaking, a pair of surfaces spanned on nearby pairs of geodesics. To be more precise, let be simplices in the layers between satisfying conditions of Definition 11.3. Moreover, assume that for each we have that or , and or . Let be associated partial characteristic discs, unique by 11.4(iv). Denote the boundary vertices of (resp. ) by (resp. ), its characteristic mapping by (resp. ).
Lemma 11.8**.**
There exists a simplicial generalized characteristic disc for and embeddings (thought of as inclusions, for simplicity) such that the distances in and distances in are all for . Moreover, for .
Proof. For each , let be the greater among and let be the smaller, let be the greater among and let be the smaller. Pick vertices so that the distance is maximal. If possible, choose them from (if it is possible for independently, then it is possible for both of them at the same time, by Lemma 9.5). Pick a 1–skeleton geodesic connecting to intersecting (this is possible by Corollary 8.7). If , then put , otherwise let be the neighbor of on . Analogically, if , then put , otherwise let be the neighbor vertex of on . Thus . Let be the partial characteristic disc for for . Denote its boundary vertices by .
The embedding, say , is defined as follows. By Proposition 7.6 there exists a characteristic surface such that . Moreover, again by Proposition 7.6, the sub–geodesic of lies in . Hence we can define the desired mapping as the composition . To check that this is an embedding it is enough to check that it preserves the the layers (Lemma 11.4(vii)) and is isometric on the layers (Lemma 11.4(iii)).
To prove the last assertion fix and assume w.l.o.g. that . Then , as desired.
Now let us prepare the statement of our final lemma. One can view it as a simple case of Theorem 10.1, case of being flat.
Let be a characteristic disc for a thick interval for the directed geodesics between and let be its diagonal, c.f. Definition 9.10. Let be simplices of the Euclidean diagonal in (Definition 9.10). Fix . If then let be directed geodesics in from to and from to respectively. If then put in the definition of and in the definition of . If then put in the definition of and in the definition of . For all other purposes we will put .
Let be the subcomplex of which is the span of the union of over all . Note that is a simplicial generalized characteristic disc. Denote the vertices of its boundary loop by and . Denote by the geodesic joining in