The Coarse Geometry of Groups
Tim SusseCUNY Graduate Center
December 2, 2011
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 1 / 38
Outline
1 Groups as Geometric Objects
2 Quasi-isometries
3 Hyperbolicity and Hyperbolic Groups
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 2 / 38
The Cayley Graph
Given a presentation of a group G = 〈S | R〉 we associate a geometricobject called the Cayley graph, denoted Cayley(G,S).
Vertex Set = Gg1 ∼ g2 if and only if there exists s ∈ S ∪ S−1 with g1 = g2s
Thus the Cayley graph of a group depends on the choice of generatingset.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 3 / 38
The Cayley Graph
Given a presentation of a group G = 〈S | R〉 we associate a geometricobject called the Cayley graph, denoted Cayley(G,S).
Vertex Set = Gg1 ∼ g2 if and only if there exists s ∈ S ∪ S−1 with g1 = g2s
Thus the Cayley graph of a group depends on the choice of generatingset.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 3 / 38
The Cayley Graph
Given a presentation of a group G = 〈S | R〉 we associate a geometricobject called the Cayley graph, denoted Cayley(G,S).
Vertex Set = Gg1 ∼ g2 if and only if there exists s ∈ S ∪ S−1 with g1 = g2s
Thus the Cayley graph of a group depends on the choice of generatingset.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 3 / 38
Cayley Graph of Z2
Figure: Cayley(Z2, {(1,0), (0,1)})
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 4 / 38
Cayley Graphs of Z/7Z
(a) (b)
Figure: (a) Cayley(Z/7Z, {1}) (b) Cayley(Z/7Z, {2,3})
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 5 / 38
Two Cayley Graphs of Z
(a)
(b)
Figure: (a) Cayley(Z, {1}) (b) Cayley(Z, {2,3})
You might notice that if you step really far back, or zoom out on thegraphic, the two Cayley graphs look very similar. We want to formalizeand take advantage of this.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 6 / 38
Two Cayley Graphs of Z
(a)
(b)
Figure: (a) Cayley(Z, {1}) (b) Cayley(Z, {2,3})
You might notice that if you step really far back, or zoom out on thegraphic, the two Cayley graphs look very similar. We want to formalizeand take advantage of this.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 6 / 38
Word Metric on a Group
Given a generating set S for G, we define the distance between twopoints in G to be their distance in the Cayley graph Cayley(G,S).
Again, this depends on the choice of generating set. This metric dS iscalled the word metric on G.
FactsFor any g ∈ G, dS(g,e) = lS(g), the word length of G.For any g,h ∈ G, dS(g,h) = lS(g−1h).G acts on (G,dS) by left multiplication, an isometry.If S is a finite generating set, closed balls are finite, so (G,dS) is aproper metric space. The converse is also true.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 7 / 38
Word Metric on a Group
Given a generating set S for G, we define the distance between twopoints in G to be their distance in the Cayley graph Cayley(G,S).
Again, this depends on the choice of generating set. This metric dS iscalled the word metric on G.
FactsFor any g ∈ G, dS(g,e) = lS(g), the word length of G.For any g,h ∈ G, dS(g,h) = lS(g−1h).G acts on (G,dS) by left multiplication, an isometry.If S is a finite generating set, closed balls are finite, so (G,dS) is aproper metric space. The converse is also true.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 7 / 38
Word Metric on a Group
Given a generating set S for G, we define the distance between twopoints in G to be their distance in the Cayley graph Cayley(G,S).
Again, this depends on the choice of generating set. This metric dS iscalled the word metric on G.
FactsFor any g ∈ G, dS(g,e) = lS(g), the word length of G.For any g,h ∈ G, dS(g,h) = lS(g−1h).
G acts on (G,dS) by left multiplication, an isometry.If S is a finite generating set, closed balls are finite, so (G,dS) is aproper metric space. The converse is also true.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 7 / 38
Word Metric on a Group
Given a generating set S for G, we define the distance between twopoints in G to be their distance in the Cayley graph Cayley(G,S).
Again, this depends on the choice of generating set. This metric dS iscalled the word metric on G.
FactsFor any g ∈ G, dS(g,e) = lS(g), the word length of G.For any g,h ∈ G, dS(g,h) = lS(g−1h).G acts on (G,dS) by left multiplication, an isometry.
If S is a finite generating set, closed balls are finite, so (G,dS) is aproper metric space. The converse is also true.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 7 / 38
Word Metric on a Group
Given a generating set S for G, we define the distance between twopoints in G to be their distance in the Cayley graph Cayley(G,S).
Again, this depends on the choice of generating set. This metric dS iscalled the word metric on G.
FactsFor any g ∈ G, dS(g,e) = lS(g), the word length of G.For any g,h ∈ G, dS(g,h) = lS(g−1h).G acts on (G,dS) by left multiplication, an isometry.If S is a finite generating set, closed balls are finite, so (G,dS) is aproper metric space. The converse is also true.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 7 / 38
Outline
1 Groups as Geometric Objects
2 Quasi-isometries
3 Hyperbolicity and Hyperbolic Groups
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 8 / 38
Bilipschitz Equivalence
Given two metric spaces (X ,dX ) and (Y ,dY ) we say that a mapf : X → Y is a k -bilipschitz map if for any pair of points x1, x2 ∈ X wehave:
1k
dX (x1, x2) ≤ dY (f (x1), f (x2)) ≤ kdX (x1, x2).
A bilipschitz map is like a stretching of the metric at every point (bybounded amounts). It is worth noting the a bilipschitz map is always atopological embedding.
A weaker notion of equivalence is a ”large scale” or coarse bilipschitzcondition. We call this a quasi-isometry.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 9 / 38
Bilipschitz Equivalence
Given two metric spaces (X ,dX ) and (Y ,dY ) we say that a mapf : X → Y is a k -bilipschitz map if for any pair of points x1, x2 ∈ X wehave:
1k
dX (x1, x2) ≤ dY (f (x1), f (x2)) ≤ kdX (x1, x2).
A bilipschitz map is like a stretching of the metric at every point (bybounded amounts). It is worth noting the a bilipschitz map is always atopological embedding.
A weaker notion of equivalence is a ”large scale” or coarse bilipschitzcondition. We call this a quasi-isometry.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 9 / 38
Bilipschitz Equivalence
Given two metric spaces (X ,dX ) and (Y ,dY ) we say that a mapf : X → Y is a k -bilipschitz map if for any pair of points x1, x2 ∈ X wehave:
1k
dX (x1, x2) ≤ dY (f (x1), f (x2)) ≤ kdX (x1, x2).
A bilipschitz map is like a stretching of the metric at every point (bybounded amounts). It is worth noting the a bilipschitz map is always atopological embedding.
A weaker notion of equivalence is a ”large scale” or coarse bilipschitzcondition. We call this a quasi-isometry.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 9 / 38
Quasi-isometric embeddings
Definition. We say that a map f : X → Y is a (k , c) quasi-isometricembedding if for every pair x1, x2 we have that
1k
dX (x1, x2)− c ≤ dY (f (x1), f (x2)) ≤ kdX (x1, x2) + c.
Further, if the map is coarsely onto (i.e. if every point in Y is distanceat most c from f (X )), we call it a quasi-isometry.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 10 / 38
Quasi-isometric embeddings
Definition. We say that a map f : X → Y is a (k , c) quasi-isometricembedding if for every pair x1, x2 we have that
1k
dX (x1, x2)− c ≤ dY (f (x1), f (x2)) ≤ kdX (x1, x2) + c.
Further, if the map is coarsely onto (i.e. if every point in Y is distanceat most c from f (X )), we call it a quasi-isometry.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 10 / 38
Quasi-isometric embeddings
Quasi-isometry defines an equivalence relation on metric spaces:Clearly the identity map is an isometry, so any space isquasi-isometric to itself.A quick computation shows that if f and g are quasi-isometries, sois f ◦ g
Symmetry is a little tricky.To show symmetry we need to construct a ”coarse inverse” of aquasi-isometry f . By this we mean a quasi-isometry f−1 so that thereexists a constant r with dX (x , f−1 ◦ f (x)) ≤ r .
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 11 / 38
Quasi-isometric embeddings
Quasi-isometry defines an equivalence relation on metric spaces:Clearly the identity map is an isometry, so any space isquasi-isometric to itself.A quick computation shows that if f and g are quasi-isometries, sois f ◦ gSymmetry is a little tricky.
To show symmetry we need to construct a ”coarse inverse” of aquasi-isometry f . By this we mean a quasi-isometry f−1 so that thereexists a constant r with dX (x , f−1 ◦ f (x)) ≤ r .
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 11 / 38
Coarse Inverse of a Quasi-isometry
To construct the inverse, we first need some quick facts.If f : X → Y is a (k , c) quasi-isometry, then it’s failure to be injective isbounded. In particular, if f (x1) = f (x2), then dX (x1, x2) ≤ kc. To seethis, look at the left-side of the definition of quasi-isometry:
1k
dX (x1, x2)− c ≤ dY (f (x1), f (x2)) = 0.
So, up to a bounded diameter ”error”, for each y ∈ f (X ), we canchoose g(y) = x , where we choose some element x ∈ f−1(y).
For each element y 6∈ f (X ), we note that there exists some y ′ ∈ f (X )with dY (y , y ′) ≤ c. Choose some y ′ with this property and let g(y) = xwhere x ∈ f−1(y).
This function is well-defined if we ignore these finite diameter”errors”. We call this sort of thing coarsely well-defined.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 12 / 38
Coarse Inverse of a Quasi-isometry
To construct the inverse, we first need some quick facts.If f : X → Y is a (k , c) quasi-isometry, then it’s failure to be injective isbounded. In particular, if f (x1) = f (x2), then dX (x1, x2) ≤ kc. To seethis, look at the left-side of the definition of quasi-isometry:
1k
dX (x1, x2)− c ≤ dY (f (x1), f (x2)) = 0.So, up to a bounded diameter ”error”, for each y ∈ f (X ), we canchoose g(y) = x , where we choose some element x ∈ f−1(y).
For each element y 6∈ f (X ), we note that there exists some y ′ ∈ f (X )with dY (y , y ′) ≤ c. Choose some y ′ with this property and let g(y) = xwhere x ∈ f−1(y).
This function is well-defined if we ignore these finite diameter”errors”. We call this sort of thing coarsely well-defined.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 12 / 38
Coarse Inverse of a Quasi-isometry
To construct the inverse, we first need some quick facts.If f : X → Y is a (k , c) quasi-isometry, then it’s failure to be injective isbounded. In particular, if f (x1) = f (x2), then dX (x1, x2) ≤ kc. To seethis, look at the left-side of the definition of quasi-isometry:
1k
dX (x1, x2)− c ≤ dY (f (x1), f (x2)) = 0.So, up to a bounded diameter ”error”, for each y ∈ f (X ), we canchoose g(y) = x , where we choose some element x ∈ f−1(y).
For each element y 6∈ f (X ), we note that there exists some y ′ ∈ f (X )with dY (y , y ′) ≤ c. Choose some y ′ with this property and let g(y) = xwhere x ∈ f−1(y).
This function is well-defined if we ignore these finite diameter”errors”. We call this sort of thing coarsely well-defined.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 12 / 38
Why is Quasi-isometry Right?
Given a finitely generated group G, there exist many finite generatingsets.
Proposition.Let S and T be two finite generating sets of a group G, then the identitymap on G is a quasi-isometry idG : Cayley(G,S)→ Cayley(G,T ).
Proof. Let S = {s1, . . . , sn} and T = {t1, . . . tm}. Since S generates Gthere exists a shortest (geodesic) spelling of each ti in the language ofS. In particular, ti = sε1
1i · · · sεkki , where εj = ±1. Let
K = max{k : lS(ti) = k}, i.e. the longest word length of a ti inCayley(G,S). Similarly, let L be the maximum word length of the si inCayley(G,T ).
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 13 / 38
Why is Quasi-isometry Right?
Given a finitely generated group G, there exist many finite generatingsets.
Proposition.Let S and T be two finite generating sets of a group G, then the identitymap on G is a quasi-isometry idG : Cayley(G,S)→ Cayley(G,T ).
Proof. Let S = {s1, . . . , sn} and T = {t1, . . . tm}. Since S generates Gthere exists a shortest (geodesic) spelling of each ti in the language ofS. In particular, ti = sε1
1i · · · sεkki , where εj = ±1. Let
K = max{k : lS(ti) = k}, i.e. the longest word length of a ti inCayley(G,S). Similarly, let L be the maximum word length of the si inCayley(G,T ).
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 13 / 38
Why is Quasi-isometry Right?
Take g ∈ G. Then say lS(g) = r . so g = s1(g) · · · sr (g), where eachsi(g) ∈ S ∪S−1. Replace each of the si with their spellings in t , and wesee that lT (g) ≤ Kr .Similarly, if lT (g) = r , then lS(g) ≤ Lr . So, for any g ∈ G we get:
1K· lT (g) ≤ lS(g) ≤ L · lT (g).
However, for any g,h ∈ G, dS(g,h) = lS(gh−1), and similarly for T . So,this turns in to a quasi-isometry. In fact, this is a bi-Lipschitzequivalence.
Notice in our Cayley graphs for Z before we had K = 3, L = 2 forS = {1} and T = {2,3}.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 14 / 38
Why is Quasi-isometry Right?
Take g ∈ G. Then say lS(g) = r . so g = s1(g) · · · sr (g), where eachsi(g) ∈ S ∪S−1. Replace each of the si with their spellings in t , and wesee that lT (g) ≤ Kr .Similarly, if lT (g) = r , then lS(g) ≤ Lr . So, for any g ∈ G we get:
1K· lT (g) ≤ lS(g) ≤ L · lT (g).
However, for any g,h ∈ G, dS(g,h) = lS(gh−1), and similarly for T . So,this turns in to a quasi-isometry. In fact, this is a bi-Lipschitzequivalence.
Notice in our Cayley graphs for Z before we had K = 3, L = 2 forS = {1} and T = {2,3}.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 14 / 38
The Milnor-Svarc Lemma
One of the fundamental tools in studying the coarse geometry ofgroups is the following fact.
Theorem.Let X be a proper metric space and let G act on X geometrically(properly discontinuously and cocompactly by isometries). Then G isfinitely generated and for a fixed x ∈ X the orbit map g 7→ gx is aquasi-isometry.
The proof involves taking a closed ball K (which is compact, since X isproper) that contains a fundamental domain for the action on X . LetS = {s ∈ G : K ∩ sK 6= ∅}. Since the action is properly discontinuous,this set is finite.
We would then show that G = 〈S〉 by making a path from x ∈ K togx ∈ gK for g ∈ G by making steps that are small enough that thecorresponding translates of K intersect. These correspond to elementsof S. In doing this, the QI bounds fall out.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 15 / 38
The Milnor-Svarc Lemma
One of the fundamental tools in studying the coarse geometry ofgroups is the following fact.
Theorem.Let X be a proper metric space and let G act on X geometrically(properly discontinuously and cocompactly by isometries). Then G isfinitely generated and for a fixed x ∈ X the orbit map g 7→ gx is aquasi-isometry.
The proof involves taking a closed ball K (which is compact, since X isproper) that contains a fundamental domain for the action on X . LetS = {s ∈ G : K ∩ sK 6= ∅}. Since the action is properly discontinuous,this set is finite.
We would then show that G = 〈S〉 by making a path from x ∈ K togx ∈ gK for g ∈ G by making steps that are small enough that thecorresponding translates of K intersect. These correspond to elementsof S. In doing this, the QI bounds fall out.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 15 / 38
The Milnor-Svarc Lemma
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 16 / 38
Consequences of the Milnor-Svarc Lemma
If G is a finitely generated group and H ≤ G with [G : H] <∞,then H y G by multiplication on the right. This action iscocompact, so H is QI to G. This is an example of what is calledcommensurability (which is stronger than QI).
If G = π1(M) where M is a compact Riemannian manifold, then Gis finitely generated and is quasi-isometric to M. For instanceπ1(S), a closed hyperbolic surface group (χ(S) < 0), isquasi-isometric to the hyperbolic plane.In a similar vein, Zn is quasi-isometric to Rn.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 17 / 38
Consequences of the Milnor-Svarc Lemma
If G is a finitely generated group and H ≤ G with [G : H] <∞,then H y G by multiplication on the right. This action iscocompact, so H is QI to G. This is an example of what is calledcommensurability (which is stronger than QI).If G = π1(M) where M is a compact Riemannian manifold, then Gis finitely generated and is quasi-isometric to M. For instanceπ1(S), a closed hyperbolic surface group (χ(S) < 0), isquasi-isometric to the hyperbolic plane.In a similar vein, Zn is quasi-isometric to Rn.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 17 / 38
QI and Algebra
There are many algebraic properties that have geometric content (i.e.are quasi-isometry invariant). In particular:
Having a finite presentation is quasi-isometry invariantHaving a finite index free subgroup is a quasi-isometry invariantHaving two topological ends is equivalent to being virtually Z, so itis a quasi-isometry invariant. Having one end is also a QIinvariant.Having a finite index nilpotent subgroup (virtual nilpotence) isequivalent to having polynomial growth (this is Gromov’sPolynomial Growth Theorem). The latter is a quasi-isometryinvariant for groups.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 18 / 38
Outline
1 Groups as Geometric Objects
2 Quasi-isometries
3 Hyperbolicity and Hyperbolic Groups
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 19 / 38
Slim Triangle Property
Defintion We say that a metric space (X ,d) is δ-hyperbolic if for anygeodesic triangle [a,b, c] we have that d([a,b], [a, c] ∪ [b, c]) ≤ δ andsimilarly for all other permutations of the letters. We call such a triangleδ-thin or δ-slim
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 20 / 38
Other Characterizations of Hyperbolicity
Hyperbolicity is innately a statement about ”large” triangles in a metricspace. In fact, any compact metric space is automatically hyperbolicwith δ equal to the (finite) diameter.
There are other, equivalent, notions of hyperbolicity.Rips proved that a space is δ-hyperbolic if and only if everyquadruple of points a,b,p ∈ X satisfy
〈b | c〉p ≥ min{〈a | b〉p , 〈a | c〉p} − δ.
〈x | y〉p =12
(d(x ,p) + d(y ,p)− d(x , y)) is called the GromovProduct.Bowditch proved in his paper on the curve complex thathyperbolicity can be formulated as a statement about coarsecenters of triangles [Bo].
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 21 / 38
Examples of Hyperbolic Metric Spaces
As the name suggests, the hyperbolic plane H, which we identifywith the complex upper half plan with the Riemannian metric
ds2 =dx2 + dy2
y2 is log(1 +√
2)-hyperbolic. Further, hyperbolic
space in any dimension is hyperbolic.
A tree (of any kind, including an R-tree) is trivially 0-hyperbolic,since any triangle is actually a tripod.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 22 / 38
Examples of Hyperbolic Metric Spaces
As the name suggests, the hyperbolic plane H, which we identifywith the complex upper half plan with the Riemannian metric
ds2 =dx2 + dy2
y2 is log(1 +√
2)-hyperbolic. Further, hyperbolic
space in any dimension is hyperbolic.
A tree (of any kind, including an R-tree) is trivially 0-hyperbolic,since any triangle is actually a tripod.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 22 / 38
Quasi-isometric Invariance
Proposition.If f : X → Y is a quasi-isometry and X is δ-hyperbolic, then Y is alsohyperbolic.
To prove this proposition, we need to study the properties ofquasi-isometric images of geodesics.
A geodesic is an isometric embedding of an interval.A (k , c)-quasigeodesic is a (k , c) quasi-isometric embedding ofan interval.Remark. If γ is a geodesic in X and f : X → Y is a(k , c)-quasi-isometric embedding, then f (γ) is a(k , c)-quasigeodesic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 23 / 38
Quasi-isometric Invariance
Proposition.If f : X → Y is a quasi-isometry and X is δ-hyperbolic, then Y is alsohyperbolic.
To prove this proposition, we need to study the properties ofquasi-isometric images of geodesics.
A geodesic is an isometric embedding of an interval.
A (k , c)-quasigeodesic is a (k , c) quasi-isometric embedding ofan interval.Remark. If γ is a geodesic in X and f : X → Y is a(k , c)-quasi-isometric embedding, then f (γ) is a(k , c)-quasigeodesic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 23 / 38
Quasi-isometric Invariance
Proposition.If f : X → Y is a quasi-isometry and X is δ-hyperbolic, then Y is alsohyperbolic.
To prove this proposition, we need to study the properties ofquasi-isometric images of geodesics.
A geodesic is an isometric embedding of an interval.A (k , c)-quasigeodesic is a (k , c) quasi-isometric embedding ofan interval.Remark. If γ is a geodesic in X and f : X → Y is a(k , c)-quasi-isometric embedding, then f (γ) is a(k , c)-quasigeodesic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 23 / 38
Quasi-isometric Invariance
Definition. A function e : N→ R is called a divergence function for ametric (length) space X if for every R, r ∈ N and any pair of geodesicsγ : [0,a]→ X and γ′ : [0,a′]→ X with γ(0) = γ′(0) = x ,R + r ≤ min{a,a′} and d(γ(R), γ′(R) ≥ e(0), any path connectingγ(R + r) to γ′(R + r) outside B(x ,R + r) must have length atleast e(r).
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 24 / 38
Quasi-isometric Invariance
In the Euclidean plane we must have that e(r) ≤ πr . Importantly,any divergence function must be linear. As you might think,divergence has to do with the size of a sphere of radius r .
In an infinite tree, the divergence is infinite, since there is only onepath between two points. This is called a ”cut point”.H2 has an exponential divergence function, this can be figured outby computing the circumference of a circle of Euclidean radius rcentered at 0 in the disc model.Is this also true in other hyperbolic spaces?
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 25 / 38
Quasi-isometric Invariance
In the Euclidean plane we must have that e(r) ≤ πr . Importantly,any divergence function must be linear. As you might think,divergence has to do with the size of a sphere of radius r .In an infinite tree, the divergence is infinite, since there is only onepath between two points. This is called a ”cut point”.H2 has an exponential divergence function, this can be figured outby computing the circumference of a circle of Euclidean radius rcentered at 0 in the disc model.Is this also true in other hyperbolic spaces?
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 25 / 38
Quasi-isometric Invariance
Theorem.If X is a δ-hyperbolic metric space, then it has an exponentialdivergence function.
Proof.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 26 / 38
Quasi-isometric Invariance
Theorem.If X is a δ-hyperbolic metric space, then it has an exponentialdivergence function.
Proof.Fix R, r ∈ N. Let γ and γ′ be two geodesics based at some point x ∈ Xwith d(γ(R), γ(R′)) > 2δ and set e(0) = 2δ.Let p be a path in X \ B(x ,R + r) from γ(R + r) to γ′(R + r). and let α∅be the geodesic from γ(R + r) to γ′(R + r). Now let m∅ be the middlepoint on the path p and let α0 be the geodesic from γ(R + r) to m∅ andα1 the geodesic from m∅ to γ′(R + r).
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 26 / 38
Quasi-isometric Invariance
Theorem.If X is a δ-hyperbolic metric space, then it has an exponentialdivergence function.
Proof.Now, for any binary string b, let mb be the midpoint of the segment of pbetween the endpoints of αb. Now let αb0 be the geodesic between thebeginning of αb and mb and αb1 the geodesic between mb and the endof αb. Keep subdividing p in this way until each segment in the divisionhas length between 1
2 and 1. If n is the number of pieces, thenlog 2l(p) ≤ n ≤ log 2l(p) + 1.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 26 / 38
Quasi-isometric Invariance
Theorem.If X is a δ-hyperbolic metric space, then it has an exponentialdivergence function.
Proof.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 26 / 38
Quasi-isometric Invariance
Theorem.If X is a δ-hyperbolic metric space, then it has an exponentialdivergence function.
Proof.For each b, the segments αb, αb0, αb1 form a geodesic triangle, and soare δ-slim.Since d(γ(R), γ′(R)) > δ, there exists a point v(0) on α∅ withd(v(0), γ(R)) < δ.Continuing inductively, we can find v(1) on α0 ∪ α1 withd(v(0), v(1)) ≤ δ. And so if v(i) is on αb we find v(i + 1) on either αb0or αb1 with d(v(i), v(i + 1)) < δ.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 26 / 38
Quasi-isometric Invariance
Theorem.If X is a δ-hyperbolic metric space, then it has an exponentialdivergence function.
Proof.Let v(m) be the point obtained at the last level of iteration. There isapoint y ∈ P whose distance from v(m) is at most 1 and so itsdistance from x is at most R + δ log2(l(p)) + 2.But d(x ,P) ≥ R + r so
R + r ≤ R + δ log2(l(p)) + 2,i.e. l(p) is atleast exponential in r .
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 26 / 38
Quasi-isometric Invariance
Proposition.Let γ be a (k , c)-quasigeodesic with end points x and y and let [x , y ]denote a geodesic (not necessarily unique) connecting x to y. Then,there exists M = M(k , c, δ) so that the Hausdorff distance between γand [x , y ] is less than M.In particular, γ is in the M-neighborhood of the geodesic between itsendpoints
The proof of this is an application of the theorem from the previousslide.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 27 / 38
Quasi-isometric Invariance
Corollary.If X is a δ-hyperbolic space and f : Y → X be a (K ,C)-quasi-isometry,then Y is K (2M + δ) + C-hyperbolic.
Proof.Let [a,b, c] be some geodesic triangle in Y and consider its imagef ([a,b, c]) in X . Well f ([a,b]) is a quasi-geodesic, so it is in theM-neighborhood of a geodesic [f (a), f (b)] and similarly for [b, c] and[a, c]. Take a point x ∈ f ([a,b]). Let y ∈ [f (a), f (b)] be such thatd(x , y) < M and let z ∈ [f (b), f (c)] ∪ [f (a), f (c)] be such thatd(y , z) < δ. Furthermore, there exists w ∈ f ([b, c]) ∪ f ([a, c]) such thatd(z,w) < M. So, d(x ,w) < 2M + δ.Since f was a quasi-isometry, and x and w are in the image of f , theirpreimages on the triangle [a,b, c] are at most K (2M + δ) + C apart.Thus the triangle in Y is slim.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 28 / 38
Quasi-isometric Invariance
Thanks to Victor Reyes for this image.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 29 / 38
Hyperbolic Groups
The previous slide shows us that it makes sense to talk about finitelygenerated hyperbolic groups. If we transition between finitegenerating sets, the Cayley graphs are quasi-isometric, so if one ishyperbolic, so is any other.
If M is a closed hyperbolic manifold (i.e. M is isometric to Hn),then π1(M) is δ-hyperbolic. (This is a consequence of theMilnor-Svarc Lemma)Given a hyperbolic group G, a subgroup H is quasiconvex if andonly if the inclusion map H ↪→ G is a quasi-isometric embedding.In a hyperbolic group, if g ∈ G is infinite order, then the centralizerof g is quasiconvex.No hyperbolic group contains as Z2 subgroup. In fact, it can notcontain a Baumslag-Solitar subgroup
⟨a,b | b−1amb = an⟩.
[Gromov] If G satisfies the small cancellation condition C′(16), then
G is hyperbolic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 30 / 38
Hyperbolic Groups
The previous slide shows us that it makes sense to talk about finitelygenerated hyperbolic groups. If we transition between finitegenerating sets, the Cayley graphs are quasi-isometric, so if one ishyperbolic, so is any other.
If M is a closed hyperbolic manifold (i.e. M is isometric to Hn),then π1(M) is δ-hyperbolic. (This is a consequence of theMilnor-Svarc Lemma)
Given a hyperbolic group G, a subgroup H is quasiconvex if andonly if the inclusion map H ↪→ G is a quasi-isometric embedding.In a hyperbolic group, if g ∈ G is infinite order, then the centralizerof g is quasiconvex.No hyperbolic group contains as Z2 subgroup. In fact, it can notcontain a Baumslag-Solitar subgroup
⟨a,b | b−1amb = an⟩.
[Gromov] If G satisfies the small cancellation condition C′(16), then
G is hyperbolic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 30 / 38
Hyperbolic Groups
The previous slide shows us that it makes sense to talk about finitelygenerated hyperbolic groups. If we transition between finitegenerating sets, the Cayley graphs are quasi-isometric, so if one ishyperbolic, so is any other.
If M is a closed hyperbolic manifold (i.e. M is isometric to Hn),then π1(M) is δ-hyperbolic. (This is a consequence of theMilnor-Svarc Lemma)Given a hyperbolic group G, a subgroup H is quasiconvex if andonly if the inclusion map H ↪→ G is a quasi-isometric embedding.In a hyperbolic group, if g ∈ G is infinite order, then the centralizerof g is quasiconvex.No hyperbolic group contains as Z2 subgroup. In fact, it can notcontain a Baumslag-Solitar subgroup
⟨a,b | b−1amb = an⟩.
[Gromov] If G satisfies the small cancellation condition C′(16), then
G is hyperbolic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 30 / 38
Hyperbolic Groups
If M is a closed hyperbolic manifold (i.e. M is isometric to Hn),then π1(M) is δ-hyperbolic. (This is a consequence of theMilnor-Svarc Lemma)Given a hyperbolic group G, a subgroup H is quasiconvex if andonly if the inclusion map H ↪→ G is a quasi-isometric embedding.In a hyperbolic group, if g ∈ G is infinite order, then the centralizerof g is quasiconvex.No hyperbolic group contains as Z2 subgroup. In fact, it can notcontain a Baumslag-Solitar subgroup
⟨a,b | b−1amb = an⟩.
[Gromov] If G satisfies the small cancellation condition C′(16), then
G is hyperbolic.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 30 / 38
Dehn Presentations
It’s natural to wonder if the word problem is solvable in a hyperbolicgroup (or any new class of groups that is defined). It turns out thathyperbolic groups have very special presentations that make the wordproblem easy.Definition. Let G = 〈S | R〉. We say that the presentation is a Dehnpresentation if for any reduced word w with w = 1 in G, there exists arelator r ∈ R so that r = r1r2, l(r1) > l(r2) and w = w1r1w2.
In other words, any word that represents the identity in G containsmore than one half of a relator, and so it can be shortened.
A word which cannot be further reduced or shortened by this method(replacing r1 by r−1
2 , a shorter word) is called Dehn reduced.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 31 / 38
Dehn Presentations
It’s natural to wonder if the word problem is solvable in a hyperbolicgroup (or any new class of groups that is defined). It turns out thathyperbolic groups have very special presentations that make the wordproblem easy.Definition. Let G = 〈S | R〉. We say that the presentation is a Dehnpresentation if for any reduced word w with w = 1 in G, there exists arelator r ∈ R so that r = r1r2, l(r1) > l(r2) and w = w1r1w2.
In other words, any word that represents the identity in G containsmore than one half of a relator, and so it can be shortened.
A word which cannot be further reduced or shortened by this method(replacing r1 by r−1
2 , a shorter word) is called Dehn reduced.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 31 / 38
Efficient Solution to the Word Problem
If G has a finite Dehn presentation (G is finitely generated, R is finite)then you can check all subwords of length at mostN = max {l(r) : r ∈ R} to see if a reduction can be made.
This procedure for solving the word problem is called Dehn’sAlgorithm, originally created by Max Dehn in 1910 to solve the wordproblem in surface groups. Its run time is O(|w |2) in its simplestiteration. (There are atmost |w | − N subwords in each step, and atmost |w | steps in the reduction.)
Example. G = 〈a,b, c,d | [a,b][c,d ]〉. Let R be the symmetrized set ofgenerators, so that R contains all cyclic conjugates of [a,b][c,d ] andits inverse. What about other hyperbolic groups?
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 32 / 38
Efficient Solution to the Word Problem
If G has a finite Dehn presentation (G is finitely generated, R is finite)then you can check all subwords of length at mostN = max {l(r) : r ∈ R} to see if a reduction can be made.
This procedure for solving the word problem is called Dehn’sAlgorithm, originally created by Max Dehn in 1910 to solve the wordproblem in surface groups. Its run time is O(|w |2) in its simplestiteration. (There are atmost |w | − N subwords in each step, and atmost |w | steps in the reduction.)
Example. G = 〈a,b, c,d | [a,b][c,d ]〉. Let R be the symmetrized set ofgenerators, so that R contains all cyclic conjugates of [a,b][c,d ] andits inverse. What about other hyperbolic groups?
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 32 / 38
Hyperbolic Groups: Local Geodesics
Definition. A path γ in a metric space X is called a k -local geodesic ifevery subpath of length k is a geodesic.
On a sphere of radius 1, a great circle is a π-local geodesic.If M is a Riemannian manifold, let r(M) be the injectivity radius ofM. Then any image of a ray in TxM under the exponential map isan r(M)-local geodesic.
Local geodesics have to do with loops in a metric space, or relations ina Cayley graph. That makes them natural to study when consideringpresentations.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 33 / 38
Hyperbolic Groups: Local Geodesics
Definition. A path γ in a metric space X is called a k -local geodesic ifevery subpath of length k is a geodesic.
On a sphere of radius 1, a great circle is a π-local geodesic.If M is a Riemannian manifold, let r(M) be the injectivity radius ofM. Then any image of a ray in TxM under the exponential map isan r(M)-local geodesic.
Local geodesics have to do with loops in a metric space, or relations ina Cayley graph. That makes them natural to study when consideringpresentations.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 33 / 38
Hyperbolic Groups: Local Geodesics
Lemma.Let G be δ-hyperbolic group and let γ be a 4δ-local geodesic. Let g bethe geodesic between the endpoints of γ (called γ+ and γ−). Assumel(g) > 2δ and let r and s be points on γ and g respectively, bothdistance 2δ from γ+. Then d(r , s) ≤ δ.
The proof here is by induction on the length of γ and uses the thintriangle property multiple times.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 34 / 38
Hyperbolic Groups: Dehn Presentation
Theorem.If γ is a 4δ-local geodesic in a δ-hyperbolic group G, then γ iscontained in the 3δ neighborhood of the geodesic between itsendpoints.
We will use this theorem to create a shortening algorithm in ourhyperbolic group G. Fix a finite generating set S. LetR = {w : w = 1 in G, |w | < 8δ}. We aim to show that 〈S | R〉 is a Dehnpresentation for G.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 35 / 38
Hyperbolic Groups: Dehn Presentation
With R as in the previous slide, take a word w in the generators so thatw = 1 in G. If the loop w in the Cayley graph is already a 4δ-localgeodesic, then it is in the 3δ neighborhood of the origin, so it is alreadyan element of R (since if it had length more than 8δ, the first 4δ longsegment of the geodesic would leave the ball of radius 3δ) and we cansee that it represents the identity.
Now say that w is not a 4δ-local geodesic. Then there exists somesubpath (subword) w1 of length 4δ which is not a geodesic between itsendpoints. Replace it in w by the geodesic between its endpoints, callit w2. Then the path w1w−1
2 has length less than 8δ, so that word is inR. Further, we note that w contained w1, the longer part of the relator.
Now, we can continue this process until we reduce w to a 4δ-localgeodesic, a case we have already covered. Thus, G = 〈S | R〉 is aDehn presentation.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 36 / 38
Hyperbolic Groups: Dehn Presentation
With R as in the previous slide, take a word w in the generators so thatw = 1 in G. If the loop w in the Cayley graph is already a 4δ-localgeodesic, then it is in the 3δ neighborhood of the origin, so it is alreadyan element of R (since if it had length more than 8δ, the first 4δ longsegment of the geodesic would leave the ball of radius 3δ) and we cansee that it represents the identity.
Now say that w is not a 4δ-local geodesic. Then there exists somesubpath (subword) w1 of length 4δ which is not a geodesic between itsendpoints. Replace it in w by the geodesic between its endpoints, callit w2. Then the path w1w−1
2 has length less than 8δ, so that word is inR. Further, we note that w contained w1, the longer part of the relator.
Now, we can continue this process until we reduce w to a 4δ-localgeodesic, a case we have already covered. Thus, G = 〈S | R〉 is aDehn presentation.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 36 / 38
Hyperbolic Groups: Dehn Presentation
With R as in the previous slide, take a word w in the generators so thatw = 1 in G. If the loop w in the Cayley graph is already a 4δ-localgeodesic, then it is in the 3δ neighborhood of the origin, so it is alreadyan element of R (since if it had length more than 8δ, the first 4δ longsegment of the geodesic would leave the ball of radius 3δ) and we cansee that it represents the identity.
Now say that w is not a 4δ-local geodesic. Then there exists somesubpath (subword) w1 of length 4δ which is not a geodesic between itsendpoints. Replace it in w by the geodesic between its endpoints, callit w2. Then the path w1w−1
2 has length less than 8δ, so that word is inR. Further, we note that w contained w1, the longer part of the relator.
Now, we can continue this process until we reduce w to a 4δ-localgeodesic, a case we have already covered. Thus, G = 〈S | R〉 is aDehn presentation.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 36 / 38
Open Questions [Be]
[Gromov] Given a hyperbolic group G with one topological end(i.e. a freely indecomposable hyperbolic group), does it contain asurface subgroup?Are hyperbolic groups residually finite?[Bestvina] Say that G admits a finite dimensional K (G,1) anddoes not contain any Baumslag-Solitar groups. Is G necessarilyhyperbolic? If G embeds in a hyperbolic group is this true? (Note:Gromov proved that every hyperbolic group admits a finitedimensional K(G, 1), making this question more natural than itseems.)[Canary] Let H ≤ G, G a hyperbolic group. If there exists some nso that gn ∈ H for every g ∈ G, is H necessarily finite index in G?(The answer is yes is H is quasiconvex)
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 37 / 38
References
[Be] Mladen Bestvina, Questions in Geometric Group Theory,http://www.math.utah.edu/ bestvina/eprints/questions-updated.pdf(2004).
[Bo] Brian Bowditch, Intersetction Numbers and Hyperbolicity of theCurve Complex, J. reine angew. Math. 598 (2008), 105-129.
[BrH] Martin Bridson and Andre Hafliger, Metric Spaces ofNon-positive Curvature: Grundlehren der mathematischenWissenschaften Series, Springer (2010).
[Gr] M. Gromov, Hyperbolic Groups: Essays in Group Theory, S. M.Gersten ed., M.S.R.I. Publ 8, Springer (1988), 75 - 263.
Tim Susse CUNY Graduate Center () The Coarse Geometry of Groups December 2, 2011 38 / 38