Determining the small solutions to 𝑆-unit equations (Q4257699)

The crucial part in solving (\(S\)-)unit equations is the final enumeration of the small solutions. The problem discussed here is as follows: Given two infinite multiplicative subgroups \(G_i\) of \(K^*\) (via a basis), numbers \(\alpha_1\), \(\alpha_2\) and bound \(H\), find all solutions \(\tau_i\in G_i\) of \(\tau_1\alpha_1 + \tau_2\alpha_2+1=0\) such that the coefficients of \(\tau_i\) with respect to the given basis are bounded by \(H\). Usually one solves this by applying sieving techniques. Since the number of possible coefficient (exponent) vectors is exponential in the \(\mathbb{Z}\)-rank of \(G_i\), this approach is limited to ``small'' groups. Recently, in his Ph. D. Thesis [``On the resolution of unit and index form equations in algebraic number fields with an application to the determination of all integer points of a Mordell curve'' (in German), Berlin: TU Berlin, FB Mathematik (1997; Zbl 0912.11061)] \textit{K. Wildanger} found a way to divide the ``solution space'' for classical unit equations geometrically into smaller parts and to show that most parts contain no solutions at all. For the remaining parts one can either use the same technique recursively or enumerate the solutions efficiently. In this paper the author extends Wildanger's method to \(S\)-unit equations. He modifies the concepts to include finite primes. The main idea is similar to Wildanger's, namely that solutions either are uniformly distant from \(1\) in valuations or at least one valuation has to be quite close to \(2\). In addition to this, the author develops strategies (based on integral LLL-reduction) to avoid real arithmetic (and therefore round off error) as long as possible, thus making the method more reliable. He demonstrates the power of this new method by discussing an \(S\)-unit equation already solved in [\textit{N. P. Smart}, Proc. Lond. Math. Soc., III. Ser. 75, No. 2, 271-307 (1997; Zbl 0885.11031)]. The old method used to take some CPU-years (i.e. weeks on a huge network) whereas the improvements reduce the runtime to some minutes.