7

P. Hall proved that every finitely generated torsion-free nilpotent group can be faithfully represented by upper unitriangular matrices over $\mathbb Z$. The most famous example is the integral Heisenberg group, which is free nilpotent, of rank $2$ and class $2$, and is isomorphic to $UT(3,\mathbb{Z})$.

I would like to know what is the easiest such representation for the free nilpotent group $F(3,2)$, of rank $3$ and class $2$. Its group presentation is $$F(3,2)=\langle x,y,z \,\|\, [x,[x,y]],[y,[x,y]],[z,[x,y]],[x,[y,z]],[y,[y,z]],[z,[y,z]],[x,[x,z]],[y,[x,z]],[z,[x,z]] \rangle.$$

An algorithm to compute this representation was implemented in GAP, based on the work of de Graaf and Nickel [de Graaf, Willem A.; Nickel, Werner, Constructing faithful representations of finitely-generated torsion-free nilpotent groups, J. Symb. Comput. 33, No. 1, 31-41 (2002). ZBL1021.20006.] Unfortunately I am not proficient with GAP to run it myself.

In the above paper the authors state that the algorithm produces a representation for $F(3,2)$ in $UT(6,\mathbb{Z})$.

Is there a faithful representation of $F(3,2)$, over integer matrices, of degree less than $6$?

  • 3
    It is easy to see that there is no such representation of degree $4$ or less, so that just leaves dimension $5$ as a candidate. I would guess that it cannot be done, but I am not sure! – Derek Holt May 05 '21 at 11:45
  • Thanks, @DerekHolt! Do you know what does the 6-dimensional representation look like? I.e., what are the images of $x,y,z$ in $UT(6,\mathbb{Z})$? – Ashot Minasyan May 05 '21 at 14:02

3 Answers3

9

I think that I have found an embedding of $F(3,2)$ into ${\rm UT}(5,{\mathbb Z})$. I used a combination of GAP and Magma to do this, but here is the (putative) result in GAP.

The images of the three generators of $F(3,2)$ are

a := [
    [ 1, 0, 1, 0, -1 ],
    [ 0, 1, 0, 0, 0 ],
    [ 0, 0, 1, 0, -1 ],
    [ 0, 0, 0, 1, 0 ],
    [ 0, 0, 0, 0, 1 ]
];;
b := [
    [ 1, 0, 0, 0, 0 ],
    [ 0, 1, 0, 1, 0 ],
    [ 0, 0, 1, 0, -1 ],
    [ 0, 0, 0, 1, 0 ],
    [ 0, 0, 0, 0, 1 ]
];;
c := [
    [ 1, 0, 0, 0, 0 ],
    [ 0, 1, 1, 1, -1 ],
    [ 0, 0, 1, 1, -1 ],
    [ 0, 0, 0, 1, 0 ],
    [ 0, 0, 0, 0, 1 ]
];;

If you calculate the three commutators $[a,b]$, $[a,c]$ and $[b,c]$, you will find that they clearly generate a subgroup isomorphic to ${\mathbb Z}^3$, and they are all centralized by $a$,$b$, and $c$, which I think is enough to show that the group they generate is isomorphic to $F(3,2)$.

gap> x:=Comm(a,b);; y := Comm(a,c);; z := Comm(b,c);;
gap> Display(x); Display(y); Display(z);
[ [   1,   0,   0,   0,  -1 ],
  [   0,   1,   0,   0,   0 ],
  [   0,   0,   1,   0,   0 ],
  [   0,   0,   0,   1,   0 ],
  [   0,   0,   0,   0,   1 ] ]
[ [   1,   0,   0,   1,  -1 ],
  [   0,   1,   0,   0,   1 ],
  [   0,   0,   1,   0,   0 ],
  [   0,   0,   0,   1,   0 ],
  [   0,   0,   0,   0,   1 ] ]
[ [  1,  0,  0,  0,  0 ],
  [  0,  1,  0,  0,  1 ],
  [  0,  0,  1,  0,  0 ],
  [  0,  0,  0,  1,  0 ],
  [  0,  0,  0,  0,  1 ] ]
Derek Holt
  • 96,726
  • Thank you, this is a great answer! Since the commutators $x,y,z$ generate a copy of $\mathbb{Z}^3$, which is central in the group, it must indeed be isomorphic to the free nilpotent group of rank 3 (because any non-trivial normal subgroup of a nilpotent group has non-trivial intersection with its centre). – Ashot Minasyan May 05 '21 at 15:34
  • Can I ask how you found these matrices? Have you simply tested all matrices in $UT(5,\mathbb{Z})$ with entries $\pm 1$? – Ashot Minasyan May 05 '21 at 17:37
  • 1
    I first found the image of the subgroup in ${\rm GL}(5,5)$. So I was looking for a subgroup of order $5^6$ with certain properties in a group of order $5^{10}$, which I did using combinations of thought and trial and error. I eventually found such a subgroup in which the matrix entries of the three generators were all $0,1$ or $-1$, so it was easy to lift it back to ${\rm GL}(5,{\mathbb Z})$. – Derek Holt May 05 '21 at 18:06
  • This is an impressive method. It's even more surprising that using a representation of a finite quotient you were able to find a faithful representation of the infinite group. Was there any particular reason why you chose to work with the prime $5$. Have you tried $2$ and $3$ first? – Ashot Minasyan May 05 '21 at 19:27
  • For primes $2$ and $3$, the finite groups are less generic, and have exponents $8$ and $9$. For $p \ge 5$, the exponent is $p$ (and the group is regular as a $p$-group). – Derek Holt May 06 '21 at 07:35
3

Here is how GAP can be used to construct a 6-dimensional faithful representation of $F(3,2)$.

gap> LoadPackage("Polycyclic");;
gap> LoadPackage("nq");;
gap> f:=FreeGroup(3);;
gap> x:=f.1;;y:=f.2;;z:=f.3;;
gap> rels:=[Comm(x,Comm(x,y)),Comm(y,Comm(x,y)),Comm(z,Comm(x,y)),Comm(x,Comm(y,z)),Comm(y,Comm(y,z)),Comm(z,Comm(y,z)),Comm(x,Comm(x,z)),Comm(y,Comm(x,z)),Comm(z,Comm(x,z))];
[ f1^-1*f2^-1*f1^-1*f2*f1*f2^-1*f1*f2, f2^-2*f1^-1*f2*f1*f2*f1^-1*f2^-1*f1*f2, f3^-1*f2^-1*f1^-1*f2*f1*f3*f1^-1*f2^-1*f1*f2, f1^-1*f3^-1*f2^-1*f3*f2*f1*f2^-1*f3^-1*f2*f3, f2^-1*f3^-1*f2^-1*f3*f2*f3^-1*f2*f3, 
f3^-2*f2^-1*f3*f2*f3*f2^-1*f3^-1*f2*f3, f1^-1*f3^-1*f1^-1*f3*f1*f3^-1*f1*f3, f2^-1*f3^-1*f1^-1*f3*f1*f2*f1^-1*f3^-1*f1*f3, f3^-2*f1^-1*f3*f1*f3*f1^-1*f3^-1*f1*f3 ]
gap> G:=f/rels;
<fp group on the generators [ f1, f2, f3 ]>
gap> p:=NilpotentQuotient(G);
Pcp-group with orders [ 0, 0, 0, 0, 0, 0 ]
gap> IsPcpGroup(p);
true
gap> rep:=UnitriangularMatrixRepresentation(p);
[ g1, g2, g3, g4, g5, g6 ] -> <6 6x6-matrices>

Here NilpotentQuotient is used just to convert a finitely presented group G into a nilpotent group, maybe there is a more straightforward way. p got 6 generators, I guess, they are the original generators and pairwise commutators (it's probably in the docs somewhere). Now one can look at the generators, e.g. the 1st and the 6th:

 gap> m:=Image(rep);
 <matrix group of size infinity with 6 generators>
 gap> mg:=GeneratorsOfGroup(m);;
 gap> Display(mg[1]);
 [ [   1,   0,   0,   0,   0,   0 ],
   [   0,   1,   0,   0,   0,   0 ],
   [   0,   0,   1,   0,   0,   0 ],
   [   0,   0,  -1,   1,   0,   0 ],
   [   0,   1,   0,   0,   1,   0 ],
   [   0,   0,   1,   0,   0,   1 ] ]
 gap> Display(mg[6]);
 [ [  1,  0,  0,  0,  0,  0 ],
   [  0,  1,  0,  0,  0,  0 ],
   [  0,  0,  1,  0,  0,  0 ],
   [  0,  0,  0,  1,  0,  0 ],
   [  0,  0,  0,  0,  1,  0 ],
   [  1,  0,  0,  0,  0,  1 ] ]

For better control, one can use explicit homomorphisms:

  gap> F:=FreeGroup(3);
  <free group on the generators [ f1, f2, f3 ]>
  gap> f:=NqEpimorphismNilpotentQuotient(F,2); # this gives F(3,2) as the image
  [ f1, f2, f3 ] -> [ g1, g2, g3 ]
  gap> Image(f);
  Pcp-group with orders [ 0, 0, 0, 0, 0, 0 ]
  gap> Image(f,F.3*F.1);
  g1*g3*g5
  gap> m:=UnitriangularMatrixRepresentation(Image(f));
  [ g1, g2, g3, g4, g5, g6 ] -> <6 6x6-matrices>
  gap> Image(m,Image(f,F.1));
  [ [ 1, 0, 0, 0, 0, 0 ], [ 0, 1, 0, 0, 0, 0 ], [ 0, 0, 1, 0, 0, 0 ], [ 0, 0, -1, 1, 0, 0 ], [ 0, 1, 0, 0, 1, 0 ], [ 0, 0, 1, 0, 0, 1 ] ]
  gap> Display(Image(m,Image(f,F.3*F.1)));
  [ [   1,   0,   0,   0,   0,   0 ],
    [   0,   1,   0,   0,   0,   0 ],
    [   0,   0,   1,   0,   0,   0 ],
    [   0,   0,  -1,   1,   0,   0 ],
    [   0,   1,   0,   0,   1,   0 ],
    [   0,   1,   1,   0,   1,   1 ] ]
  • Thank you, this is very useful. Is there a command to see what the images of the original 3 generators $f_1,f_2,f_3$ are under this representation? Also, what does the Nilpotent Quotient command actually do? Can you apply this command only to a nilpotent group, or to any group? – Ashot Minasyan May 06 '21 at 08:16
  • 2
    I've added to my answer details on how to actually get matrix images of words in the original generators. Indeed, NilpotentQuotient computes the "biggest", in a naturally defined sense, quotient group (up to a fixed class, 2nd parameter) of an arbitrary f.p. group. More details on the algorithm and the references to theory, see https://www.gap-system.org/Manuals/pkg/nq/doc/chap0.html – Dima Pasechnik May 06 '21 at 09:49
  • In GAP 4.12dev with nq 2.5.4 I get error messages at NilpotentQuotient(G), like The 1st argument is 'fail' which might point to an earlier problem among other things. Maybe you have encountered such errors in similar cases and know what to do? – მამუკა ჯიბლაძე Aug 26 '21 at 08:24
  • please file a bug report on https://github.com/gap-packages/nq – Dima Pasechnik Aug 26 '21 at 14:53
1

The group $F(r,c)$ embeds in the $\mathbf{Q}$-group associated to the free $c$-step nilpotent Lie $\mathbf{Q}$-algebra on $r$ generators. The latter possesses a natural grading in $\{1,\dots,r\}$.

More generally if a finite-dimensional Lie algebra $g$ over a field $K$ is graded in $\mathbf{Z}$ with $g_0=\{0\}$ [this can be shown to imply $g$ nilpotent], then it has a natural faithful representation in dimension $\dim(g)+1$.

Indeed, first define $D$ as the (invertible) derivation of $g$ acting as $\times k$ on $g_k$. Let $h$ be the semidirect product $K\ltimes g$, where $1\in K$ acts as $D$. Then $h$ has trivial center, so its adjoint representation is faithful, hence is faithful in restriction to $g$. With a little effort and exponentiation one should be able to provide explicit matrices for the corresponding representation of $F(r,c)$.

This is the easiest faithful representation I can figure out, which doesn't mean it's the easiest. And doesn't mean it's the smallest one (indeed for $F(3,2)$ this yields dimension 7, while other answers provide a 5-dimensional one — for $F(r,2)$ it yields dimension $1+r+r(r-1)/2$).

YCor
  • 18,715
  • Thanks, Yves! I was thinking that an explicit faithful representations of $F(3,2)$ in $UT(n,\mathbb{Z})$ must be somewhere in the literature, but I have not found it. The representation discovered by @DerekHolt worked great for me. – Ashot Minasyan May 10 '21 at 08:40
  • 2
    @AshotMinasyan actually it's been done for all nilpotent Lie algebras of dimension $\le 6$ over an arbitrary field of char 0 (or large finite char.) [this specific one is denoted $L_{6,26}$ there]. See N. Rojas, Minimal representations for 6-dimensional nilpotent Lie algebra, J. Algebra Appl. Vol. 15, No. 10 (2016) 1650191 (19 pages) https://ri.conicet.gov.ar/bitstream/handle/11336/59910/CONICET_Digital_Nro.a86e2a8b-5b9c-4ce1-9c69-8c6dd522e722_A.pdf?sequence=2 – YCor May 10 '21 at 08:54