BOUNDED GENERATION BY SEMI-SIMPLE ELEMENTS: QUANTITATIVE

RESULTS

PIETRO CORVAJA, JULIAN L. DEMEIO, ANDREI S. RAPINCHUK, JINBO REN, AND UMBERTO M. ZANNIER

Abstract. We prove that for a number field F , the distribution of the points of a set Σ ⊂ An F with a
arXiv:2203.00755v1 [math.NT] 1 Mar 2022

purely exponential parametrization, for example a set of matrices boundedly generated by semi-simple
(diagonalizable) elements, is of at most logarithmic size when ordered by height. As a consequence, one
obtains that a linear group Γ ⊂ GLn (K) over a field K of characteristic zero admits a purely exponential
parametrization if and only if it is finitely generated and the connected component of its Zariski closure
is a torus. Our results are obtained via a key inequality about the heights of minimal m-tuples for
purely exponential parametrizations. One main ingredient of our proof is Evertse’s strengthening of
the S-Unit Equation Theorem.

1. Statement of Main Results

The purpose of this note is to annouce and sketch certain results in a future paper by the current
authors [4].
We start by the notion of bounded generation. An abstract group Γ is said to have the bounded
generation property (BG) if it can be written in the form
Γ = hγ1 i · · · hγr i
for certain fixed γ1 , . . . , γr ∈ Γ, where hγi i is the cyclic subgroup generated by γi . We refer the
interested readers to the discussion in Section 1 of [3] and the references therein for the motivation
for (BG). In [3], it was shown that a linear group Γ ⊂ GLn (K) over a field K of characteristic zero
“usually lacks (BG) by semi-simple elements”, i.e. (BG) such that all γi are diagonalizable. More
precisely, it was shown in [3] that if a linear group Γ over a field of characteristic zero consists entirely
of semi-simple elements, then Γ has (BG) if and only if it is finitely generated and virtually abelian.
In particular, if K is a number field and S is a finite set of places including all infinite ones, then
infinite S-arithmetic subgroups of absolutely almost simple K-anisotropic groups never have (BG).
The current paper will significantly strengthen the above results by providing some quantitative
properties which describe the extent of the absence of (BG) by semi-simple elements. In fact, we will
consider the following more flexible question in terms of purely exponential polynomial parametriza-
tions (PEP).
Definition. Let Σ be a subset of a variety V ⊂ AnK (K is a field). Then Σ is said to have Purely
Exponential Parametrization (PEP) in r variables if Σ has shape
n o
Σ = (f1 (n), . . . , fs (n)); n ∈ Zr ,
where each fi (x) = f (x1 , . . . , xr ) is a Purely Exponential Polynomial, i.e. an expression of the
form
e
l (x) l (x)
X
fi (x) = aj λ11,j · · · λkk,j ,
j=1
×
for certain constants a1 , . . . , ae , λ1 , . . . , λk ∈ K and linear forms lj,s (x) in r variables whose co-
efficients are rational integers. Here we refer to the elements λ1 , . . . , λk as the bases of f : =
(f1 , . . . , fs ), to the linear forms li,j as the exponents of f , and to the constants aj as the coeffi-
cients of f .
Remark 1.1. In the definition as above, we do not require that all coefficiens and bases are in K.
Also, it is easy to see that any finite union of (PEP) sets is still a (PEP) set.
1
2 P. CORVAJA, J. DEMEIO, A. RAPINCHUK, J. REN, AND U. ZANNIER

Example 1.2. The classical Pell equations naturally produce (PEP) sets. For example, the set of
integer solutions of x2 − 2y 2 = 1, which corresponds to the integer points of the special orthogonal
group for the quadratic form h = x2 − 2y 2 , is given by
( √ n √ n! √ n √ n !! )
(3 − 2 2) + (3 + 2 2) (3 − 2 2) − (3 + 2 2)
(−1)m , √ ; m, n ∈ Z .
2 2 2
Example 1.3. Linear groups Γ admitting (BG) by semi-simple elements, which are main study objects
of [3], become typical examples of (PEP) sets. In fact, if Σ = Γ ⊂ GLn (K) with Γ = hγ1 i · · · hγr i with
the γi ’s semi-simple, then there exist gi ∈ GLn (K) and λi,j for i = 1, . . . , r, j = 1, . . . , n with
gi−1 γi gi = diag(λi,1 , . . . , λi,n ), for all i = 1, . . . , r.
This implies that every γ ∈ Γ has shape
r h i
gi diag(λai,1i , . . . , λai,n
Y
γ= i
) gi−1 for some a1 , . . . , ar ∈ Z.
i=1
2
Comparing entries of the two sides of the above relation, we realize Σ as a (PEP) set ⊂ AnK in r
variables with bases equal to those eigenvlues λi,j ’s.
In the current article, we will provide some sparseness results for (PEP) subsets of affine varieties
V ⊂ AnK over a number field K. The language we are using to describe sparseness is the height
function on the affine space K n , defined by
 Y 1/[K : Q]
Haff (x1 , . . . , xn ) : = H(1 : x1 : · · · : xn ) : = max{1, kx1 kv , . . . , kxn kv }
v∈VK

where VK is the set of all places of K, and k · kv are normalized v-adic valuations such that the
product formula holds. We will also use the corresponding logarithmic height haff : = log Haff . See
[11, §B] or [2] for details about height functions. The first main result of this paper, which is about
the distribution of (PEP) sets, can be stated as follows.
Theorem 1.4 (First Main Theorem: quantitative result). Let AnK be an affine space over a number
field K, then for any (PEP) set Σ ⊂ AnK in r variables, we have

{P ∈ Σ; Haff (P ) ≤ H} = O((log H)r ) when H → ∞.

In other words, any (PEP) set has at most logarithmic-to-the-r growth in terms of the height.
Remark 1.5. In order to interprete Theorem 1.4 as a sparseness result, it should be emphasized that
there is a highly involved but also well-developed topic about “counting lattice points in Lie groups”. In
particular, [12, Corollary 1.1] (see also [9, Theorem 2.7 and Theorem 7.4]) informs us that points in any
lattice of a non-compact semi-simple Lie group G with finite center have growth rate cH d (log H)e , H →
2
∞ for certain c, d > 0, e ≥ 0 in terms of an Euclidean norm on Rn ⊃ GLn (R) ⊃ G. As a consequence,
we see that for a semi-simple algebraic group G ⊂ GLn over Q of non-compact type, Theorem 1.4
provides sparseness, in terms of the height, for all (PEP) subsets of Γ : = G(Z) : = GLn (Z) ∩ G(R).
As a more explicit example, according to [6, Example 1.6], the set {s ∈ SLn (Z); Haff (s) ≤ H} is of
2
order cH n −n for some c > 0, therefore any (PEP) set in Γ = SLn (Z), which has only logarithmic
growth, is sparse in terms of the height. Verification of sparseness for (PEP) subsets of many other
S-arithmetic groups, following strategies developed in [9] and [10], will be available in [4].
If we apply Theorem 1.4 to the particular situation of (BG) by semi-simple elements, we acquire
the following consequence.
Corollary 1.6. Let Γ ⊂ GLn (K) be a linear group over a number field K, then for any semi-simple
elements γ1 , . . . , γr ∈ Γ, we have
{P ∈ hγ1 i · · · hγr i; Haff (P ) ≤ H} = O((log H)r ) when H → ∞.
 BOUNDED GENERATION BY SEMI-SIMPLE ELEMENTS: QUANTITATIVE RESULTS 3

The proof of Theorem 1.4 relies crucially on a key statement about the so-called “minimal m-tuples”
with respect to a (PEP) set which seems to be of independent interest.
Definition 1.7. Given a vector f = (f1 , . . . , fs ) of exponential polynomials in r variables, i.e. each
fj is an exponential polynomial in r variables, an element n = (n1 , . . . , nr ) ∈ Zr is called f -minimal
(or minimal with respect to f ) if for all n′ = (n′1 , . . . , n′r ) ∈ Zm with f (n′ ) = f (n) (i.e. fj (n′ ) = fj (n)
for all j), we have kn′ k∞ : = max{|n′1 |, . . . , |n′r |} ≥ max{|n1 |, . . . , |nr |} = : knk∞ .
Theorem 1.8 (Primary Height Inequality). Let f be a vector of purely exponential polynomials in r
variables, then there exists a constant C = C(f ) > 0 such that for all f -minimal vectors n ∈ Zr , we
have

(1) haff (f (n)) ≥ C · knk∞

except on some set of the form f −1 (A) with A finite.
It should be emphasized that the constant C above will be explicitly computable, while the cardi-
nality of the set A in Theorem 1.8 is non-effective in general, see Remark 2.4.
The first main Theorem, i.e. Theorem 1.4, being quantitative itself, leads us to the following
qualitative theorem which fully describes all linear groups admitting (BG) by semi-simple elements
(or (PEP)). It is worth pointing out that, thanks to a specialization argument, the following result
works for linear groups over arbitrary fields of characteristic zero.
Theorem 1.9 (Second Main Theorem: qualitative result). Let K be a field of characteristic zero and
let Γ ⊂ GLn (K) be a linear group. Then the following three properties are equivalent.
(1) Γ has (PEP).
(2) Γ consists only of semi-simple elements and has (BG).
(3) Γ is finitely generated and the connected component G◦ of the Zariski closure G of Γ is a torus
(in particular, Γ is virtually abelian).
This result serves as an extension of one main Theorem in [3, Theorem 1.1] which claims that if a
linear group over a field of characteristic zero has (BG) by semi-simple elements, then it is virtually
solvable. More importantly, Theorem 1.9 gives a complete answer to the Questions asked in [3, p.
3].

2. Brief outline of proofs

It is straightforward to verify that Theorem 1.8 implies Theorem 1.4. For simplicity of argument,
we only sketch the proof of Theorem 1.8 for f = f being a single purely exponential polynomial. The
sketch we give here follows the lines of the proof in the general case, and already includes all the main
ideas and ingredients in the counterpart in [4].

Sketch of proof of Theorem 1.8. The proof goes by induction on r, the number of variables in n. The
base case when r = 0 is trivial, now let r ≥ 1. We write:
e
X
f (n) = ai ui (n),
i=1

l (n ,...,n )
r l (n ,...,n )
r
where λj , ai ∈ K ∗ and ui (n) = λ11,i 1 · · · λkk,i 1 are purely exponential monomials.
Some non-trivial but routine manipulations enable one to reduce to the case where λ1 , . . . , λk are
multiplicatevely independent, i.e. λθ11 · · · λθkk = 1(θj ∈ Z) ⇐⇒ θ1 = · · · θk = 0, and where the linear
forms li,j span the dual space of Qr over Q.
We need the following crucial height inequality which can be derived from a result of Evertse [8,
Theorem 6.1.1] (which is itself a consequence of the Schlickewei-Schmidt Subspace Theorem, cf. [5,
Theorem 2.2]).
4 P. CORVAJA, J. DEMEIO, A. RAPINCHUK, J. REN, AND U. ZANNIER

Theorem 2.1 (Evertse). Let S be a finite set of places of a number field K containing all archimedean
ones. Then there exists an effective C > 0 such that the inequality
haff (s1 + · · · + se ) < C · (haff (s1 ) + · · · + haff (se )) with si ∈ OS×
has only finitely many solutions such that the sum s1 + · · · + se is non-degenerate.
P
Here non-degenerate refers to the fact that i∈I si 6= 0 for any nonempty proper subset I ⊆
{1, . . . , e}.
Using Theorem 2.1 by taking the set S of places such that all bases λi and coefficients aj of f are
S-units, we obtain that for certain C ′ > 0, the inequality:
(2) haff (f (n)) ≥ C ′ · (haff (u1 (n)) + . . . + haff (ue (n)))
holds for all but finitely many n ∈ Zr such that the sum defining f (n) is non-degenerate.
Recall the following standard fact [17, p.118, Eq. (3.12)]:
Proposition 2.2. Let m ∈ N and φ = (φ1 , . . . , φe ) : Zr → (K ∗ )e be an injective group homomorphism.
Then there are constants C2 > C1 > 0 such that for every n ∈ Zr the following inequalities hold:
C1 knk∞ ≤ haff (φ1 (n)) + . . . + haff (φe (n)) ≤ C2 knk∞ .
Using the proposition above with φ = (u1 , . . . , ue ), which is injective because of the assumption
that the linear forms li,j span (Qr )∨ and that those λj ’s are multiplicatively independent, one deduces
that the right hand side of (2) is ≍ knk∞ . This completes the argument for the non-degenerate case.
Now consider those n ∈ Zr such that the sum defining f (n) is degenerate. Then we may take a
proper subset I = {i1 , . . . , it } ⊂ {1, . . . , e}(t < e) with ai1 ui1 (n) + · · · + ait uit (n) = 0.
We are now in a position to use Laurent’s theorem [8, Theorem 10.10.1], which can also be deduced
from Theorem 2.1.
Theorem 2.3 (Laurent). Let K be a number field, Γ ⊆ (K ∗ )t be a finitely generated subgroup, and
let X be a subvariety of (Gm )t . Then the Zariski closure of Γ ∩ X is a finite union of cosets of (Gm )t .
Employing Laurent’s theorem on the subgroup Γ = im(φ = (ui1 , . . . , uit ) : Zr → (K ∗ )t ) and the
hyperplane X : ai1 xi1 + . . . + ait xit = 0, and letting I go through all proper subsets of {1, . . . , e}
(finitely many possibilities), we deduce that the set of such n is contained in a finite union of cosets
of Zr . Moreover, due to the assumption that the linear forms li,j span (Qr )∨ , we may assume these
cosets are all translates of subgroups of rank < r.
Taking the restriction of f to each of the above proper cosets, and composing it with a suitable affine
transformation, we produce finitely many (PEP) sets whose parametrizations all involve < r variables.
Applying the induction hypothesis to these new (PEP) sets, the proof is completed. 
Remark 2.4 (effectiveness). Taking more care in the proof above, one can actually make the constant
C in Theorem 1.8 to be effectively computable in terms of f .
However, our approach can say little about the effectiveness of the exceptional set A (and even
less about the effectiveness of f −1 (A)) of Theorem 1.8, not even its cardinality. As a consequence,
in the context of Theorem 1.4, we are unable to explicitly compute a constant a > 0 such that
{P ∈ Σ; Haff (P ) ≤ H} < a · (log H)r for sufficiently large H.
This is in sharp contrast with the situation of S-unit equations, e.g. x1 + · · · + xs = 1, xi ∈ OS∗ ,
whose number of non-degenerate solutions can be effectively boundable from above, cf. the seminal
paper [7] and its refinement [1], see also [16] for another approach.
In fact, we prove in [4] that an effective bound for the cardinality of A in Theorem 1.8 would yield
an explicit bound for the size of non-degenerate solutions to an arbitrary S-unit equation, which is
still an open problem. Thus, the non-effectiveness of the exceptional set A of Theorem 1.8 lies deeply
in the openness of the difficult effectiveness problem of the Schlickewei-Schmidt subspace theorem.
We now turn to the discussion of the second main result, Theorem 1.9. The proof of Theorem 1.9,
being non-trivial though, is roughly analogous to that of Theorem 1.1 and Corollary 1.2 of [3]. In
particular, the theory of generic elements, cf. [13],[14], [15], will be needed again. We will omit the
 BOUNDED GENERATION BY SEMI-SIMPLE ELEMENTS: QUANTITATIVE RESULTS 5

full verification here for simplicity of presentation. In the following we will only highlight two new
ingredients in the proof of Theorem 1.9 and postpone detailed arguments in [4].
The first one is a consequence of Theorem 1.4.
Corollary 2.5. Let K be a number field, Σ ⊆ GLn (K) be a (PEP) subset, and let g ∈ GLn (K) be a
non-semi-simple matrix. Then there is an m ∈ N such that {n ∈ N; n ≤ N and g n ∈ Σ} = O(logm N )
as N → ∞ .
Proof. Write g = gu gs for the Jordan decomposition of g with gu unipotent, gs semisimple and [gu , gs ] =
1. Note that the condition gn = (gu gs )n ∈ Σ implies that gun ∈ Σ · hgs i, and that the subset Σ′ =
Σ · hgs i ⊆ GLn (K) is also a (PEP) set. So, we reduce to proving the result for gu . We may, therefore,
assume that g is unipotent.
By writing g = id+gN with gN nilpotent, and considering the binomial expansion of gn = (id + gN )n ,
it is easy to check that the height of the coefficients of gn has polynomial growth in n. Due to Theorem
1.4, the elements of height ≤ H in the (PEP) set Σ grow at most as some power of log H as H → ∞.
This proves the corollary. 
The following second requires a not entirely trivial argument which uses the finiteness of non-
degenerate solutions to S-unit equations (cf. [4]).
Lemma 2.6. Let f : Zr → K ∗ be a (PEP). If its image is a multiplicative subgroup of K ∗ , then this
subgroup is finitely generated.
Details of the proofs in this section as well as relevant examples and remarks will appear in [4].
Acknowledgements. The first author is partially funded by the Italian PRIN 2017 “Geometric, algebraic and
analytic methods in arithmetic”. The second author was a guest at the Max Planck Institute for Mathematics
when working on this article. He thanks the Institute for their hospitality and their financial support. The
fourth author is supported by the Institute for Advanced Study and the National Science Foundation under
Grant No. DMS-1926686.

References
1. F. Amoroso, E. Viada, Small points on subvarieties of a torus, Duke Math. J. 150, (2009), no. 3, p. 407-442.
2. E. Bombieri, W. Gubler, Heights in Diophantine Geometry, Cambridge University Press 2006.
3. P. Corvaja, A. Rapinchuk, J. Ren, U. Zannier, Non-virtually abelian anisotropic linear groups are not boundedly
generated, Invent. Math. 227 (2022), no. 1, 1-26.
4. P. Corvaja, J. Demeio, A. Rapinchuk, J. Ren, U. Zannier, Sparseness of purely exponential parametrization sets
and its applications, in preparation.
5. P. Corvaja, U. Zannier, Applications of Diophantine approximation to integral points and transcendence, Cambridge
Tracts in Mathematics 212, Cambridge Univ. Press, 2018.
6. W. Duke, Z. Rudnick, P. Sarnak, Density of integer points on affine homogeneous varieties, Duke Math.J. 71 (1993),
no. 1, 143-179.
7. J. H. Evertse, H. P. Schlickewei, W. Schmidt, Linear equations in variables which lie in a multiplicative group, Ann.
of Math. (2) 15) (2002), no. 3, p. 807-836.
8. J.H. Evertse, K. Győry, Unit Equations in Diophantine Number Theory, Cambridge studies in advanced mathematics
146, Cambridge University Press 2015.
9. A. Gorodnik, B. Weiss, Distribution of lattice orbits on homogeneous varieties, Geom. Funct. Anal. 17 (2007), no.1,
p. 58-115.
10. A. Gorodnik, A. Nevo, Counting lattice points, J. Reine Angew. Math. 663 (2012), p. 127-176.
11. M. Hindry, J. H. Silverman, Diophantine Geometry: An Introduction, GTM 201, Springer, 2000.
12. F. Maucourant, Homogeneous asymptotic limits of Haar measures of semisimple linear groups and their lattices,
Duke Math. J. 136 (2007), no. 2, p. 357-399.
13. G. Prasad, A.S. Rapinchuk, Existence of irreducible R-regular elements in Zariski-dense subgroups, Math. Res. Lett.
10(2003), 21-32.
14. G. Prasad, A.S. Rapinchuk, Generic elements in Zariski-dense subgroups and isospectral locally symmetric spaces,
Thin subgroups and superstrong approximation, 211-252, Math. Sci. Res. Inst. Publ. 61, Cambridge Univ. Press,
2014.
15. G. Prasad, A.S. Rapinchuk, Generic elements of a Zariski-dense subgroup form an open subset, Trans. Moscow
Math. Soc. 78(2017), 299-314.
16. G. Rémond, Sur les sous-variétés des tores, Compositio Math. 143 (2002), no. 3, p. 337-366.
6 P. CORVAJA, J. DEMEIO, A. RAPINCHUK, J. REN, AND U. ZANNIER

17. U. Zannier, Lecture notes on Diophantine analysis, Appunti. Scuola Normale Superiore di Pisa (Nuova Serie)
[Lecture Notes. Scuola Normale Superiore di Pisa (New Series)], vol. 8, Edizioni della Normale, Pisa, 2009, With
an appendix by Francesco Amoroso, xvi+237 pages.

Dipartimento di Scienze Matematiche, Informatiche e Fisiche, via delle Scienze, 206, 33100 Udine, Italy
Email address: [email protected]

Max Planck Institute for Mathematics, Bonn, 53111, Germany

Email address: [email protected]

Department of Mathematics, University of Virginia, Charlottesville, VA 22904-4137, USA

Email address: [email protected]

School of Mathematics, Institute for Advanced Study, Princeton, NJ 08540, USA

Email address: [email protected]

Scuola Normale Superiore, Piazza dei Cavalieri, 7, 56126 Pisa, Italy

Email address: [email protected]