On the number of terms of a power of a polynomial (Q5919977)

For a polynomial \(f\) (over an arbitrary field \(k\)) denote by \(t(f)\) the number of its non-zero coefficients. It has been conjectured by \textit{A. Rényi} [Acta Math. 1, 30-34 (1947; Zbl 0030.11402)] and \textit{P. Erdős} [Nieuw Arch. Wisk. (2) 23, 63--65 (1949; Zbl 0032.00203)] that for complex polynomials a bound for \(t(f^2)\) implies a bound for \(t(f)\). This was established by the first author in 1987 [Acta Arith., 49, 55--70 (1987; Zbl 0632.12024)] in a more general form, replacing the square by an arbitrary power. Now the authors show that if \(l\) is a positive integer and either char \(k=0\), or char \(k>l\deg f\), then \(t(f^l)\geq2+(\log(t(f))-1)/\log(4l)\), and the same inequality holds if char \(k>l^{t(f^l)-1}(t(f)^2-t(f)+2)\).