hot topics
DESCRIPTION
Talk at the MSRI conference in hot topics in thin groups and super-strong approximation on February 10 2012.TRANSCRIPT
![Page 1: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/1.jpg)
Some generic properties of some occasionally
obscure groups Igor Rivin, Temple University
MSRI Hot Topics Thin Groups and Super Strong Approximation
![Page 2: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/2.jpg)
In the beginning...
Thurston’s philosophy: everything is hyperbolic.
(Actually Poincare’s philosophy, adopted by Thurston).
![Page 3: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/3.jpg)
Examples:Poincare (?): Almost all oriented surfaces are hyperbolic.
Thurston: Almost all Dehn surgeries on a knot are hyperbolic.
Thurston: Almost all surface bundles over the circle are hyperbolic ???
![Page 4: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/4.jpg)
The last one seems
reasonable...
Thurston: A surface bundle (viewed as a mapping torus of a map F) is hyperbolic if and only if F is “pseudo-anosov”.
![Page 5: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/5.jpg)
And
Surface automorphisms, are periodic, reducible, or pseudo-anosov (Thurston’s classification), so seems reasonable that a random element is pseudo-anosov.
What is “random”?
![Page 6: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/6.jpg)
Two interpretations (for a discrete group):
Pick a (symmetric) generating set, look at long random products.
Look at elements in a Cayley ball.
![Page 7: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/7.jpg)
Third interpretation for
matrix groups
Look at all elements of bounded “height” (for example, matrix norm).
![Page 8: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/8.jpg)
But
It took a while to make it precise. More precisely:
![Page 9: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/9.jpg)
It never rains but when it pours
(2006)
![Page 10: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/10.jpg)
I. Kapovich
Suggested that it would be easier to translate the problem to the symplectic group (the Torelli map sends the mapping class group of genus g to Sp(2g, Z), and is surjective -- see Farb-Margalit).
![Page 11: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/11.jpg)
Key ingredientCasson’s criterion: A map F is pseudo-anosov if P = characteristic polynomial of T(F) -- has the following propertie:
P is irreducible.
P is not cyclotomic.
P is not of the form
![Page 12: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/12.jpg)
What interpretation of
randomness do we use?
Random walk or Cayley ball? Or “Archimedean height”?
Yes to all three, sort of.
![Page 13: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/13.jpg)
Graph walk
Put generators of the group onto vertices of a (Perron-Frobenius, undirected) graph.
Allow multiplication if there is an edge.
![Page 14: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/14.jpg)
“Stupid” random walk
Corresponds to the complete graph.
![Page 15: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/15.jpg)
Cayley graph
Corresponds to walk on the defining automaton (group has to be bi-automatic).
![Page 16: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/16.jpg)
Proof
In brief, a form of sieving, using ideas on equidistribution of walks on finite groups, and the already mentioned Borel-Chavdarov result (IR 2007, published DMJ ’08, FM ’09), and property tau
![Page 17: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/17.jpg)
Results (SL(2, Z) Hua-Reiner)
![Page 18: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/18.jpg)
Results (SL(3, Z) Hua-Reiner)
![Page 19: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/19.jpg)
Results (SL(3, Z) transvections)
![Page 20: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/20.jpg)
Results (SL(4, Z) Hua-Reiner)
![Page 21: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/21.jpg)
Archimedean height
Completely different, we can do it anyway (using Nevo-Sarnak (Acta, 2010), see also Gorodnik-Nevo’s book and papers).
![Page 22: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/22.jpg)
The method also works for Out(Fn)
There, the result is that a random outer automorphism is an IWIP (irreducible with irreducible powers).
Need the Galois group to be not intransitive, so just show it is the full symmetric group (generalized by Kowalski et al).
![Page 23: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/23.jpg)
Argument is good
Because gives completely effective bounds (in terms of property tau constants, which are effective as per Kassabov, Shalom), exponential convergence.
![Page 24: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/24.jpg)
Argument is bad
Because only seems to be for finite index subgroups of the mapping class group
In particular, no result at all for the Torelli subgroup (kernel of T).
![Page 25: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/25.jpg)
Maher’s result is the opposite:
Not effective
Very general (works for any “nonelementary” subgroup of the mapping class group).
Argument geometric, so does not seem to work for Out(Fn)
![Page 26: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/26.jpg)
More Maher
Maher’s argument also gives growth of the translation length in the curve complex.
Which shows that a random Dunfield-Thurston manifold is hyperbolic.
![Page 27: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/27.jpg)
What is the state of the art for the
MCG?Bestvina-Fujiwara (2008) gave ineffective but general result via quasimorphisms.
Malyutin (2010), using the Bjorkman-Hartnick central limit theorem showed (by soft methods) that the convergence in BF is superpolynomial (but can’t show exponential).
![Page 28: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/28.jpg)
State of the art continued.
Joseph Maher can get exponential convergence (still ineffective?) 2011
In his original paper also shows increasing distance in the curve complex (also follows from BF).
![Page 29: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/29.jpg)
State of the art, continued more
(2011)Lubotzky-Meiri and Malestein-Souto can do Torelli, by a “Prym map” trick (going to double covers, and getting a linear representation of Torelli, an idea of Looijenga, and of Grunewald-Lubotzky).
Lubotzky-Meiri can do Torelli of Out(Fn)
![Page 30: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/30.jpg)
Question 1:
Can you do sieving directly on the mapping class group (using CSP a la Ellenberg)?
![Page 31: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/31.jpg)
Back to our scheduled programSuperstrong approximation
allows us to catch up a little to Maher, since we can prove all of our results for thin subgroups of linear groups.
But also, we can show that a random subgroup (of a Zariski dense subgroup) is Zariski dense (IR 2010, R. Aoun 2011)
![Page 32: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/32.jpg)
And also
Generic subgroup of a linear group is free (R. Aoun 2011).
Generic subgroup of a word-hyperbolic group is free (Gilman-Miasnikov-Osin).
![Page 33: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/33.jpg)
Combining the last couple of slides
Generic subgroup (in higher rank, anyhow) is thin.
![Page 34: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/34.jpg)
Very thin
In fact, for a two-generator subgroup with random long-word generators, Hausdorff dimension goes to zero (Fuchs-IR, 2012)
![Page 35: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/35.jpg)
And yetA generic (2-generator) subgroup of a linear group of rank > 1 is profinitely dense (modular reductions are ALL surjective) with positive probability. For SL(2, Z) need 3 random generators. (Capdebosq-IR, 2012).
![Page 36: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/36.jpg)
Historical note
First example of a free profinitely dense subgroup was constructed by Stephen Humphreys (’87), second by Soifer-Venkataramana (’00), independently, various other people probably also.
![Page 37: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/37.jpg)
So, thin groups are good
Because there are lots of them.
Because they are easy to verify: Lubotzky, Weigel: surjective for one prime implies p-adic density implies Zariski density (as pointed out in A. Rapinchuk’s talk).
![Page 38: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/38.jpg)
Well, actually...
Not clear how easy...
![Page 39: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/39.jpg)
Question 2:
Given matrix generators for a subgroup G of (say) SL(n, Z), can one give a bound for how far a certificate of Zariski-density or lack thereof lies? (in terms of the matrix coefficients)
![Page 40: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/40.jpg)
Question 3
Same question as Question 2, but for profinite density.
![Page 41: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/41.jpg)
Thin is bad
Since, as far as I know, the bounds are not very effective (theoretically effective?) unlike lattices.
![Page 42: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/42.jpg)
Lattices are good
Because get good bounds
![Page 43: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/43.jpg)
Lattices are bad
Because thin on the ground, and it seems very difficult to check if something is a lattice.
![Page 44: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/44.jpg)
Model question:
Given a set of matrices A, B, C..., do they generate SL(n, Z)?
![Page 45: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/45.jpg)
Easier questionDoes <A, B, C, ...> contain M?
Undecidable for SL(n>3, Z)! (Mihailova, 195?) -- uses F2xF2, shows equivalence to Post correspondence.
Open(!!!) for SL(3, Z)
Easy for SL(2, Z)
![Page 46: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/46.jpg)
Back to Model question
The Mihailova result would seem to suggest that the model question is undecidable also, but this is open, as far as I know.
![Page 47: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/47.jpg)
What we really want to know is...whether H=<A, B, C, ...> is of
finite index.
This is roughly equivalent to the model question by the congruence subgroup property (it is usually not difficult to tell which congruence subgroup H would be if it were a congruence subgroup...), then check if H generates the congruence subgroup.
![Page 48: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/48.jpg)
Question 4:
In our applications, we have two-generator subgroups. Is that any easier than the general case? That’s canonical, since:
![Page 49: Hot topics](https://reader035.vdocument.in/reader035/viewer/2022062319/5559c5b3d8b42aaa6f8b5348/html5/thumbnails/49.jpg)
Question 4, continued:
Sharma-Venkataramana: every lattice contains a three-generator lattice, so three is probably as hard as the general case.
One generator seems tractable...