A remark on the Dirichlet values of a completely reducible polynomial \(\pmod p\) (Q1148349)

Let \(q\) and \(p\) be primes such that \(p\equiv 1 \pmod q\), and let \(\chi\) be a Dirichlet character of order \(q \pmod p\). Let \[ f(x) = (x + a_1)^{w_1}(x + a_2)^{w_2} \cdots (x + a_n)^{w_n} \] be a polynomial with integral roots satisfying \(0\le a_1<a_2 < \ldots <a_n < p\), and with the non-negative integers \(w_i\) not divisible by \(q\). Write \(l = a_n - a_1\). The author shows that: Theorem. Let \(p > (8l^3)^l\). Then there is a positive integer \(t\le (8l^3)^l\) such that, if \(\chi(f(n))\) has a constant value on the interval \((A, A+H] \subset (0, p - a_n)\), then there is a polynomial \(\psi\) of degree \(\le l\) so that \[ \chi(n ) = \exp \left(2\pi iq^{-1}\psi(n)\right) \] holds over the interval \(t^{-1}B <n\le t^{-1}(B +H)\), \(B = \vert A\vert +a_1\). A consequence is: Corollary. If \(\chi(f(x))\) is constant on the interval \((0,H] \subset (0,p - a_n)\), and \(H\ge (a_n+ 1)^3 (8l^3)^{l+1}\) then \(\chi(n) = 1\) on \((0, (8l^3)^{-l-1}H)\). This is superior to earlier similar results of the reviewer [Proc. Lond. Math. Soc., III. Ser. 13, 537--548 (1963; Zbl 0118.04704)] and of \textit{N. M. Stephens} [Bull. Lond. Math. Soc. 11, 52--54 (1979; Zbl 0412.10029)] in that it is uniform in \(q\). These earlier results were based on showing that the hypothesis \[ \chi(f(n))\text{ is constant on }(0,H] \tag{1} \] implied that there existed a \(b\) (bounded in terms of \(f\) and \(g\)) and such that, if \(H>b\), \[ \chi(n) = \chi(n +1)\text{ on }(0,H]. \tag{2} \] The argument of Stephens required \(H>q^{2a_r}\), while that of the reviewer did not state an explicit lower bound for \(H\). The author introduces \(E\) as the shift operator \(E(h(n)) = h(n +1)\). With this notation (1) and (2) may be written respectively as \[ \left( \sum_{j=1}^n w_jE^{a_j}\right) \operatorname{ind} n \pmod q \text{ is constant on }(0,H], \] and \[ (E-1) \operatorname{ind} n \pmod q \text{ is constant on }(0,H].\] But now from field theory it follows that \(\sum_{j=1}^n w_jE^{a_j}\) divides a polynomial of the form \(E^{a_1} (E^t - 1)^l\),for some \(t\le (8l^3)^l\), and this implies \[ E^{a_1} (E^t - 1)^l \operatorname{ind} n \pmod q \text{ is constant on }(0,H]. \] From this the author deduces (2) under his weaker conditions on \(H\).