Counting Integral Points on Complete Intersections via Iterated Differencing
For a system of forms of equal degree defining a non-singular complete intersection over the rationals, the number of integral zeros of height at most B is expected to obey an asymptotic formula of Hardy-Littlewood type, with a leading term given by a product of local densities. Iterating a multidimensional q-analogue of van der Corput differencing to arbitrary depth, rather than stopping after a single application, converts information about the dimension of the singular locus of the system into bounds on the associated exponential sums. This yields the expected asymptotic under a codimension hypothesis that grows more slowly in the degree than the classical requirement, with a genuine gain only once the degree is at least four, since only then can the differencing be iterated more than once.
Diophantine equations in moderately many variables
Given polynomials f1,...,fr in Z[x1,...,xn] of degrees d1,...,dr, let N(f,B) count the integer points x with |x| <= B at which all r polynomials vanish simultaneously. When the polynomials cut out a …