The Banach-Tarski Paradox, the von Neumann-Dayconjecture
Cornelia Drutu
Oxford
TCC Course 2019, Lecture 4
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 1 / 23
Assessment
Exercise Sheet 1 is now on the webpage
http://people.maths.ox.ac.uk/drutu/tcc6/
For any type of assessment (broadening courses or others) answers mustbe sent in by email at [email protected] before June 12.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 2 / 23
Last lecture
Last lecture I
Gromov’s Question: When are two quasi-isometric spaces actuallybi-Lipschitz equivalent? Especially if the spaces are separated.Key examples: Finitely generated groups, separated nets in Euclideanspaces.
Burago-Kleiner, McMullen: examples of separated nets in R2 notbi-Lipschitz equivalent.T. Dymarz: examples of lamplighter groups (which are amenablegroups) quasi-isometric, not bi-Lipschitz equivalent.
K. Whyte: A quasi-isometry between two non-amenable finitelygenerated groups is at bounded distance from a bi-Lipschitz map.More generally true for two non-amenable graphs of boundedgeometry.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 3 / 23
Last lecture
Last lecture II
A group is amenable if there exists a mean m on `∞(G ) invariant byleft multiplication.Equivalently, if there exists a finitely additive probability measure µ onP(G ), the set of all subsets of G , invariant by left multiplication.
Invariant by left multiplication (left-invariance) can be replaced byright-invariance or by bi-invariance.
A group is amenable iff one (every) Cayley graph of G is amenable.A Cayley graph G is amenable: ∃ a Følner sequence (Ωn) ⊂ G .
For every A ⊂ G define
µn(A) =|A ∩ Ωn||Ωn|
.
|µn(A)− µn(Ag)| ≤ 2∂V (Ωn)|Ωn| when g ∈ S .
Let ω be a non-principal ultrafilter on N.Take µ(A) = ω-limµn(A) .
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 4 / 23
Last lecture
Last lecture III
A finitely generated group is either paradoxical or amenable.
One can define amenability for groups that are not finitely generated,using an invariant mean or probability measure.
The metric definition is no longer equivalent.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 5 / 23
Group operations, more examples
Group operations
Proposition
A subgroup of an amenable group is amenable.
Proof: Let H ≤ G and µ be G -invariant finitely additive probabilitymeasure on P(G ).
Axiom of Choice: ∃ subset D of G intersecting each right coset Hg inexactly one point.
Define ν(A) := µ(AD) left-invariant f.a.p. measure on H.
Corollary
Any group containing a free non-abelian subgroup is non-amenable.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 6 / 23
Group operations, more examples
1 A finite extension of an amenable group is amenable.
2 Let N be a normal subgroup of a group G . The group G is amenableif and only if both N and G/N are amenable.
3 The direct limit G of a directed system (Hi )i∈I of amenable groupsHi , is amenable.
Corollary
A group G is amenable if and only if all finitely generated subgroups of Gare amenable.
Proof of (2). µG/N(A) = µG (π−1(A)), where π : G → G/N.Let ν be a left-invariant f.a.p. measure on G/N, and λ a left-invariantf.a.p. measure on N.On every left coset gN define a f.a.p. measure λg (A) = λ(g−1A).For every subset B in G define µ(B) =
∫G/N λg (B ∩ gN)dν(gN) .
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 7 / 23
Group operations, more examples
Solvable groups
Corollary
Every solvable group (not necessarily finitely generated) is amenable.
G ′ = derived subgroup [G ,G ] of the group G .The iterated commutator subgroups G (k) defined inductively by:
G (0) = G ,G (1) = G ′, . . . ,G (k+1) =(G (k)
)′, . . .
The derived series of G is
G D G ′ D . . .D G (k) D G (k+1) D . . .
G is solvable if there exists k such that G (k) = 1.The minimal k such that G (k) = 1 is the derived length of G .
Example: The group of upper triangular n × n matrices in GL(n,K ), Kfield, is solvable.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 8 / 23
Group operations, more examples
Solvable groups continued
Milnor-Wolf: A finitely generated solvable group is either virtuallynilpotent or of exponential growth.
A group G is virtually (***) if it has a finite index subgroup with property(***).
Exercise
Suppose G is direct limit of Gi , i ∈ I . Assume that there exist k ,m ∈ N sothat for every i ∈ I , the group Gi contains a solvable subgroup Hi of index6 k and derived length 6 m. Then G contains a subgroup H of index 6 kand derived length 6 m.
Back to
Corollary
Every solvable group is amenable.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 9 / 23
Group operations, more examples
Elementary amenable
Definition (M. Day)
The class of elementary amenable groups EA= the smallest classcontaining all finite groups, all abelian groups and closed under finite-indexextensions, direct limits, subgroups, quotients and extensions
1→ G1 → G2 → G3 → 1,
where both G1,G3 are elementary amenable.
There exist groups of intermediate growth (hence amenable) but notelementary amenable: Grigorchuk groups
There exist elementary amenable groups not virtually solvable (moreoverfinitely presented): the Houghton’s groups Hn, satisfying
1→ Σ→ Hn → Zn−1
where Σ is the group of permutations with finite support of Z.See Lee, Sang Rae, Geometry of Houghton’s groups, PhD Dissertation,University of Oklahoma.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 10 / 23
Group operations, more examples
An alternative for EA
Theorem (Chou)
A finitely generated elementary amenable group either is virtually nilpotentor it contains a free non-abelian subsemigroup.
Theorem (Chou)
The class of elementary amenable groups is the smallest class containingall finite groups, all abelian groups and closed under direct limits andextensions
1→ G1 → G2 → G3 → 1,
where both G1,G3 are in the class.
Reference: Ching Chou, Elementary amenable groups, Illinois J. Math. 24(1980).
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 11 / 23
Von Neuman-Day conjecture
We are now able to relate amenable groups to the Banach–Tarski paradox.
Proposition
1 The group of isometries Isom(Rn) with n = 1, 2 is amenable.
2 The group of isometries Isom(Rn) with n > 3 is non-amenable.
Does there exist a purely algebraic definition of amenability for groups?
Conjecture
Does every non-amenable group contain a free non-abelian subgroup?
The question is implicit in von Neumann’s initial paper (1929), formulatedexplicitly by Day in 1957.
Counter-examples:
Al. Olshanskii (1980);
S, Adyan (1982): the free Burnside group B(n,m) with n ≥ 2 andm ≥ 665, m odd.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 12 / 23
Von Neuman-Day conjecture
Positive answers
Theorem (Jacques Tits 1972)
A subgroup G of GL(n,F ), where F is a field of zero characteristic, iseither virtually solvable or it contains a free nonabelian subgroup.
Also true for fields of positive characteristic, if G finitely generated.
Theorem (Mostow–Tits)
A discrete amenable subgroup G of a Lie group L with finitely manycomponents, contains a polycyclic group of index at most η(L).
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 13 / 23
Von Neuman-Day conjecture
Other classes of groups for which the von Neumann-Day conjecture has apositive answer (in fact Tits’ theorem is true):
1 subgroups of Gromov hyperbolic groups (Gromov);
2 subgroups of the mapping class group of a surface (Ivanov 1992);
3 subgroups of Out(Fn) (Bestvina-Feighn-Handel 2000,2004,2005);
4 fundamental groups of compact manifolds of nonpositive curvature(Ballmann 1995).
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 14 / 23
Von Neuman-Day conjecture
A metric von Neumann-Day
A metric version of the von Neumann-Day conjecture established byBenjamini and Schramm:
A locally finite graph G with positive Cheeger constant contains a treewith positive Cheeger constant.Uniform bound on the valency is not assumed.
Cheeger constant is considered with edge–boundary.
If, moreover, the Cheeger constant of G is at least an integer n > 0,then G contains a spanning subgraph, with each connectedcomponent is a rooted tree with all vertices of valency n, except theroot, of valency n + 1.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 15 / 23
Von Neuman-Day conjecture
Metric von Neumann-Day continued
If X is either a graph or a Riemannian manifold with infinitediameter, bounded geometry and positive Cheeger constant (inparticular, if X is the Cayley graph of a paradoxical group) then Xcontains a bi-Lipschitz embedding of the binary rooted tree.
Related to the above, the following is asked:
Open question
(Benjamini-Schramm 1997) Does every Cayley graph of every finitelygenerated group with exponential growth contain a tree with positiveCheeger constant?
Open case: amenable non-linear groups with exponential growth.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 16 / 23
Von Neuman-Day conjecture
Next, we shall overview briefly quantitative approaches to:
non-amenability: Tarski numbers for groups;
amenability: uniform amenability and Følner functions.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 17 / 23
Quantitative non-amenability
Quantitative non-amenability
One can measure “how paradoxical” a group G is via the Tarski number.
In this discussion, groups are not required to be finitely generated.
Recall that a paradoxical decomposition of a group G is a partition
G = X1 t ... t Xk t Y1 t ... t Ym
for which ∃ g1, ..., gk , h1, ..., hm in G , so that
g1 X1 t ... t gk Xk = G
andh1 Y1 t ... t hm Ym = G .
The Tarski number of the decomposition is k + m.
The Tarski number Tar(G ) of the group = minimum of the Tarskinumbers of paradoxical decompositions.
If G is amenable then set Tar(G ) =∞.Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 18 / 23
Quantitative non-amenability
Tarski numbers and group operations
Proposition
1 Tar(G ) ≥ 4 for every group G .
2 If H 6 G then Tar(G ) 6 Tar(H).
3 Tar(G ) = 4 if and only if G contains a free non-abelian sub-group.
4 Every paradoxical group G contains a finitely generated subgroup Hwith Tar(G ) generators, such that Tar(G ) = Tar(H).
5 If N is a normal subgroup of G then Tar(G ) 6 Tar(G/N).
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 19 / 23
Quantitative non-amenability
Co-embeddable groups
Two groups G1 and G2 are co-embeddable if there exist injective grouphomomorphisms G1 → G2 and G2 → G1.
1 All countable free groups are co-embeddable.
2 Sirvanjan-Adyan: for every odd m > 665, two free Burnside groupsB(n;m) and B(k;m) of exponent m, with n > 2 and k > 2, areco-embeddable.
G1 (non-)amenable iff G2 (non-)amenable. Moreover Tar(G1) = Tar(G2).
Consequence: For every odd m ≥ 665, and n ≥ 2, the Tarski number ofB(n;m) is independent of the number of generators.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 20 / 23
Quantitative non-amenability
Paradoxical decomposition and torsion
Proposition
1 If G admits a paradoxical decomposition
G = X1 t X2 t Y1 t . . . t Ym,
then G contains an element of infinite order.
2 If G is a torsion group then Tar(G ) ≥ 6.
The Tarski numbers help to classify the groups non-amenable and withoutan F2 subgroup (“infinite monsters”).
Ceccherini, Grigorchuk, de la Harpe: The Tarski number of a free Burnsidegroup B(n;m) with n > 2 and m > 665, m odd, is at most 14.
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 21 / 23
Quantitative non-amenability
Tarski numbers, final
We proved that a paradoxical group G contains a finitely generatedsubgroup H with Tar(G ) generators, such that Tar(G ) = Tar(H).
Consequence: if G is such that all m generated subgroups are amenablethen Tar(G ) ≥ m + 1.
M. Ershov: certain Golod-Shafarevich groups G
have an infinite quotient with property (T);
for every m large enough, G contains finite index subgroups Hm withthe property that all their m-generated subgroups are finite.
Consequences:
the set of Tarski numbers is unbounded;
Tarski numbers, when large, are not quasi-isometry invariants. Noteven commensurability invariants.
Ershov-Golan-Sapir: D. Osin’s torsion groups have Tarski number 6.Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 22 / 23
Quantitative non-amenability
Questions
Question
How does the Tarski number of a free Burnside group B(n;m) depend onthe exponent m? What are its possible values?
Question
Is the Tarski number of groups a quasi-isometry invariant, when it takessmall values?
For Tar(G ) = 4 this question is equivalent to a well-known open problem.
A group G is small if it contains no free nonabelian subgroups. Thus, G issmall iff Tar(G ) > 4.
Question
Is smallness invariant under quasi-isometries of finitely generated groups?
Cornelia Drutu (Oxford) Banach-Tarski, von Neumann-Day TCC Course 2019, Lecture 4 23 / 23