Linear equations in variables which lie in a multiplicative group. (English) Zbl 1026.11038

Let \(K\) be a field of characteristic 0 and let \(n\) be a natural number. Let \(\Gamma\) be a subgroup of the multiplicative group \((K^*)^m\) of finite rank \(r\). Given \(a_1,\dots,a_n\in K^*\), let \(A(a_1,\dots,a_n,\Gamma)\) be the number of solutions \(x = (x_1,\dots,x_n)\in \Gamma\) of the equation \(a_1x_1 + \dots + a_nx_n = 1\) such that no proper subsum of \(a_1x_1 + \dots + a_nx_n\) vanishes. The main result of the paper is \[ A(a_1,\dots,a_n,\Gamma) \leq A(n,r) = \exp((6n)^{3n}(r+1)). \] As remarked by the authors, the essential feature of their inequality is uniformity, since \(A(n,r)\) is independent of \(K\) and of the coefficients \(a_1,\dots,a_n\). Moreover, although the upper bound \(A(n,r)\) is probably far from best possible, a number of examples show that a certain dependence on \(n\) and \(r\) is needed.
In particular cases, even better upper bounds for \(A(a_1,\dots,a_n,\Gamma)\) had already been proved: for \(n=2\) one has the estimate \( A(a_1,a_2,\Gamma) \leq 2^{9(r+1)}\) [F. Beukers and H. P. Schlickewei, Acta Arith. 78, 189-199 (1996; Zbl 0880.11034)] and for \(r=0\) and arbitrary \(n\) one has the estimate \(A(a_1,\dots,a_n,\Gamma) \leq (n+1)^{3(n+1)^2}\) [J.-H. Evertse, Acta Arith. 89, 45-51 (1999; Zbl 0974.11012)]. However, for all other cases the result is new.
Also, for \(K\) a number field, \(\Gamma\) may be viewed as a subgroup of the group of \(S\)-units of \(K\) (where \(S\) is a finite set of places of \(K\) including the Archimedean places), and the finiteness of \(A(a_1,\dots,a_n,\Gamma)\) had already been established (the first result is due to Mahler for \(n=2\)). An explicit upper bound was found by Schlickewei and Schmidt, but this upper bound depended not only on \(n\) and \(r\) but also on the degree \(d=[K:{\mathbb Q}]\) of the number field \(K\). The result of the paper finally solves the problem of finding a bound independent of \(d\).
In a second theorem, the authors apply their main result to linear recurrence sequences of order \(n\). It is known after Skolem-Mahler-Lech that, if \(u_m\) is a linear recurrence sequence, the solutions \(k\) of \(u_k=0\) are contained in a finite set of arithmetical progressions plus a finite set. The bound for the number of arithmetical progressions plus the number of points proved in the paper is \(\exp((6n)^{3n})\), hence only doubly exponential in \(n\), improving on an estimate of Schmidt, which is triply exponential.
The proofs use mainly two tools: the quantitative version of the Subspace Theorem, proved by Evertse and Schlickewei, to deal with large solutions and a method of Schmidt, involving lower bounds for the height of points in algebraic varieties, to deal with small solutions.


11D45 Counting solutions of Diophantine equations
11D04 Linear Diophantine equations
Full Text: DOI arXiv