computing with matrix groups, or "how dense is dense"

48
How dense is dense, or: Computing in linear groups Igor Rivin (Temple University)

Upload: igor-rivin

Post on 19-Jul-2015

161 views

Category:

Science


0 download

TRANSCRIPT

Page 1: Computing with matrix groups, or "how dense is dense"

How dense is dense, or:

Computing in linear groupsIgor Rivin

(Temple University)

Page 2: Computing with matrix groups, or "how dense is dense"

Computing in matrix

groups

One goal is to understand examples.

Another is to think of the computational aspects

as another facet of the mathematics.

Page 3: Computing with matrix groups, or "how dense is dense"

Computing in matrix

group

So, we want to develop quick practical methods

on the one hand.

And show the EXISTENCE of efficient

algorithms on the other.

If we are lucky, the two things are the same.

Page 4: Computing with matrix groups, or "how dense is dense"

Basic question: given a collection of matrices

A, B, C, D, …, what sort of group do they

generate?

Page 5: Computing with matrix groups, or "how dense is dense"

Easy version: the matrices live in a

finite group (Say, SL(n, Z/pZ))

Page 6: Computing with matrix groups, or "how dense is dense"

Hard version: the matrices live in,

say, SL(n, Z)

Page 7: Computing with matrix groups, or "how dense is dense"

The Easy version:

Given a collection of elements in, say SL(n,

Z/pZ), we can tell exactly what subgroup they

generate (Neumann-Praeger, and others) in

(probabilistic) polynomial time.

Page 8: Computing with matrix groups, or "how dense is dense"

The hard version:

Given a collection of matrices in SL(n, Z), what

can we say?

Page 9: Computing with matrix groups, or "how dense is dense"

Some history

Page 10: Computing with matrix groups, or "how dense is dense"

Once upon a time

People understood a lot about lattices.

And they could use sophisticated analysis to

answer questions about them.

Page 11: Computing with matrix groups, or "how dense is dense"

Then, came the Apollonian packing

(well, really, the work of Graham, Lagarias, Mallows, Wilks, Yan)

Page 12: Computing with matrix groups, or "how dense is dense"

Where the group is thin

Page 13: Computing with matrix groups, or "how dense is dense"

And nothing was ever the

same

Page 14: Computing with matrix groups, or "how dense is dense"

Super-strong approximation

machine

The Sarnak School (Sarnak, Helfgott, Bourgain,

Gamburd, Kontorovich, Fuchs, Salehi-Golsefidy,

Varju) and the non-Princeton school (Breuillard,

Green, Guralnick, Tao, Pyber, Szabo…)

developed a machine which allowed us to

extend the results on lattices to thin groups.

Page 15: Computing with matrix groups, or "how dense is dense"

But there were questions

Such as: other than the Apollonian packings, do

thin groups actually arise in nature?

Page 16: Computing with matrix groups, or "how dense is dense"

What is nature?

I had shown that a random subgroup of a linear

group was Zariski dense (’11), and R. Aoun

(slightly later) showed that a random subgroup

was free, so together these results showed that

a random subgroup was thin.

Page 17: Computing with matrix groups, or "how dense is dense"

But…

That is not the same as arising in nature (for

example, most numbers are transcendental, but

most numbers we run into are not…)

Page 18: Computing with matrix groups, or "how dense is dense"

So, what is nature?

Monodromy groups of families?

Page 19: Computing with matrix groups, or "how dense is dense"

Monodromies

Algebraic geometers mostly cared about Zariski

closure, but still, there was one case (due to

A’Campo) where something was shown to be

arithmetic, and a couple of cases (Deligne-

Mostow, M. Nori) which were NOT arithmetic

(but in a product).

Page 20: Computing with matrix groups, or "how dense is dense"

Calabi-Yau

Then, we (= Elena Fuchs, Inna Capdeboscq,

Sarnak, IR) saw the explicit examples of

monodromy associated to Calabi-Yau three-

folds (14 in all), and the question was: are they

thin or arithmetic?

Page 21: Computing with matrix groups, or "how dense is dense"

Are Calabi-Yaus thin?

We had computers, but it was not clear what to

do with them - we realized that no algorithms

existed, and the questions were probably

undecidable.

Page 22: Computing with matrix groups, or "how dense is dense"

But then they were

decided!

C. Brav and H. Thomas showed that seven of

the groups were thin (by showing that they

played ping-pong), and T. N. Venkataramana

and Singh showed that the other 7 were

arithmetic (by finding many unipotent).

Page 23: Computing with matrix groups, or "how dense is dense"

Still, this actually required thought, not just

CPU cycles (though Brav/Thomas used those

too).

Page 24: Computing with matrix groups, or "how dense is dense"

So, the questions

Given a collection of matrices in (say) SL(n, Z):

Is the group they generate FINITE?

Is their span Zariski dense?

If yes, is it maximal?

If not, what’s the closure?

Is their span Arithmetic?

Is their span PROFINITELY dense?

If arithmetic, what is the index?

Page 25: Computing with matrix groups, or "how dense is dense"

Finiteness

Very well understood: a number of different practical

algorithms (Babai, Babai-Rockmore, Detinko-Flannery-

O’Brien)

Basic idea of (one half of) Babai’s algorithm: If the group

is finite, then trace is bounded by N (the dimension).

Look at long product: if trace is bounded, probably finite,

otherwise not.

Page 26: Computing with matrix groups, or "how dense is dense"

Finiteness

Basic idea of Detinko-Flannery-O’Brien: look at

the intersection with a principal congruence

subgroup (this can be computed): that is torsion

free, so should be trivial.

If it is, the congruence homomorphism is an

isomorphism.

Page 27: Computing with matrix groups, or "how dense is dense"

Finiteness

Both algorithms are both theoretically and

practically good (Babai’s implemented in GAP,

DFO in MAGMA), the DFO algorithm works for

any characteristic 0 ground field.

Page 28: Computing with matrix groups, or "how dense is dense"

Zariski density

Three years ago, no one knew a good algorithm

(or at least admitted to it).

Now there are several.

Page 29: Computing with matrix groups, or "how dense is dense"

Zariski Density: Algorithm

1

Based on strong approximation: The main

observation is a theorem of T. Weigel: If some

modular projection is surjective (for p> 3), then

the group is Zariski-dense (and converse is

strong approximation)

Page 30: Computing with matrix groups, or "how dense is dense"

Zariski Density: Algorithm

1

So, in practice, pick a moderately large prime p,

reduce mod p, use Neumann-Praeger to see if

onto.

What if the answer is “NOT ONTO”?

Page 31: Computing with matrix groups, or "how dense is dense"

Zariski density: Algorithm

1

In practice: try another random prime, then quit.

In theory: Use E. Breuillard’s bound, reduce mod a prime

bigger than his bound, Zariski-dense if and only if the

projection is onto.

Problem: we know that this is a polynomial algorithm, but

we don’t know the constants.

Page 32: Computing with matrix groups, or "how dense is dense"

Zariski density: Algorithm

2

Group is Zariski-dense if and only if the adjoint

representation is irreducible and does NOT have

finite image.

Page 33: Computing with matrix groups, or "how dense is dense"

Zariski Density: Algorithm

2

We know how to check finiteness, for

irreducibility use Burnside (the group should

span the matrix algebra): polynomial time! Bad

degree (as function of dimension)! (best bound I

know is 14, so not so great in any reasonable

dimension).

Page 34: Computing with matrix groups, or "how dense is dense"

Zariski density: Algorithm

3 (IR)

Fact 1: (Prasad-Rapinchuk) If you have two non-commuting

elements in G, one of which has Galois group (of char. poly)

equal to the Weyl group of the ambient group, then G is

Zariski dense.

Fact 2: (IR, Jouve-Kowalski-Zywina) a long word in the

generators of a Z. Dense subgroup has Galois group the Weyl

group with probability exponentially close (in length) to 1.

Page 35: Computing with matrix groups, or "how dense is dense"

Zariski Density: Algorithm

3(IR)

So, algorithm is: compute two long words. If they

commute, NO. If they don’t commute, compute

Galois group (of one of them). If same as Weyl

group. YES, otherwise, NO.

Page 36: Computing with matrix groups, or "how dense is dense"

Zariski density: Algorithm

3

Problem: exponent in exponential convergence

is NOT effective (since based on super-strong

approximation).

Lesser problem: how to compute Galois group?

Page 37: Computing with matrix groups, or "how dense is dense"

Zariski Density: Algorithm

3

Weyl groups are usually the symmetric group, or

the signed permutation group - turns out that

there are (with some major caveats) good

algorithms.

Page 38: Computing with matrix groups, or "how dense is dense"

Zariski Density: Algorithm

3

For example, a polynomial time (but not practical) algorithm to check that the

Galois group is the symmetric group is to check that the characteristic polynomial

of the first five exterior powers of the companion matrix are irreducible (and the

discriminant of the original polynomial is square free).

Running time: polynomial of degree around 40(!)

We use a different algorithm, to get the running time of Zariski-density checker to

a fourth degree polynomial in the dimension for SL(n, Z), and an eighth degree

polynomial for Sp(2n, Z) (the running time is linear in the log of the height of the

generateng set, in both cases)

Page 39: Computing with matrix groups, or "how dense is dense"

What if not Zariski

dense?

An algorithm to compute Zariski closure (using

Groebner bases) is given by Derksen-Jeandel-

Koiran (’07). Not obviously practical (worst case

at least doubly exponential), but there may be

ways to make it so over Q.

Page 40: Computing with matrix groups, or "how dense is dense"

Profinite Density

Fact: a random group is profinitely dense with

probability bounded away from zero (Capdeboscq-

IR, ’15), but how do you tell?

Only algorithm I know: check every prime (primes

and 4 and 9 are enough) until Breuillard’s bound, so

simply exponential. Can one do better?

Page 41: Computing with matrix groups, or "how dense is dense"

Arithmeticity

Three years ago, we had no clue: it was easy to see

that seeing if you have the whole group was semi-

decidable (keep multiplying until you get the

generators).

Since then: Detinko-Flannery-Hulpke gave an

algorithm to compute the index of an arithmetic

subgroup.

Page 42: Computing with matrix groups, or "how dense is dense"

Arithmeticity

Their algorithm should be a semi-decision procedure: if you

let it run, it will give you either the index or a lower bound

on index (together with “don’t know”). But it is not (yet).

In particular, the arithmetic Calabi-Yau monodromies (as

described in Venkataramana’s talk) can not yet be detected

WITHOUT thinking!

Page 43: Computing with matrix groups, or "how dense is dense"

Arithmeticity (slides borrowed from

Alla Detinko)

As a finale, we show some results:

Page 44: Computing with matrix groups, or "how dense is dense"
Page 45: Computing with matrix groups, or "how dense is dense"
Page 46: Computing with matrix groups, or "how dense is dense"

Calabi-Yau (from preprint of

Hoffman-van Straten)

Page 47: Computing with matrix groups, or "how dense is dense"

Index of arithmetic

subgroups

Note: the algorithm is (obviously) practical, but

no complexity bounds are known.

Page 48: Computing with matrix groups, or "how dense is dense"

Index of arithmetic

subgroups

The work of Detinko/Flannery/Hulpke reduces

the question of “Is a subgroup finite index in a

given arithmetic group?” to “Does a given set of

matrices generate a given arithmetic group?”