On local representation of zero by a form (Q2399698)

A fundamental question in algebraic number theory is whether a homogenous form of some degree \(d\) in \(k\) variables, with integer coefficients, represents 0 non-trivially. The main obstruction is to find whether it represents 0 over the \(p\)-adic integers for every prime \(p\). The Hasse-Minkowski Theorem states that for \(n=2\) and \(k=5\) this is always the case (but not when \(k=4\)), which lead Artin to conjecture that for fixed \(n\), any form in \(n^{2}+1\) variables represents 0 over \(\mathbb{Z}_{p}\). This was proven to hold for \(n=3\), but in [C. R. Acad. Sci., Paris, Sér. A 262, 612 (1966; Zbl 0133.29705)] \textit{G. Terjanian} constructed a form of degree 4 in 18 variables that does not represent 0 over \(\mathbb{Z}_{2}\), thus disproving the conjecture. Therefore it became interesting to find weaker lower bounds for \(k\) for which the statement will still hold. The current paper is a report on a result of the author with Arkhipov, stating that the bound must be exponential in \(n\). More precisely, fix an integer \(r\), and then for making sure that any form of degree \(n\) for some \(n>n_{0}(r)\) in \(k\) variables represents 0 over the \(p\)-adic numbers, one must take \(k \geq e^{u}\) when \(\frac{n}{u}=\prod_{t=1}^{r}\ln^{(t)}n\cdot(\ln^{(r+1)}n)^{3}\) (where \(\ln^{(t)}\) is the composition of \(\ln\) with itself \(t\) times). The way of doing so is by constructing a form in enough variables that cannot represent 0. In more detail, a form is said to \textit{trivially represent 0 modulo \(p\)} if wherever its value is 0 modulo some higher power of \(p\), all the variables must be divisible by \(p\) (such a form clearly cannot represent 0). The form of Terjanian is obtained by a form in 3 variables that takes only few possible values modulo 16, and using six copies of it to get the form in 18 variables. The current paper follows these lines, but is technically harder, because the key lemma involves solving several congruences together. However, a smart interplay between the degree and the number of variables yields, in several iterations, the desired result in an elementary manner (requiring only congruences modulo prime powers). The proof is given for \(p=2\), but is stated to work in the same manner for any \(p\).