Finite-dimensional Mahler measure of a polynomial and Szegö's theorem (Q676235)

If \(P(z) = a_n \prod_{i=1}^n (z- \alpha_i)\) is a polynomial with integer coefficients, then the Mahler measure of \(P\) is defined by \[ M(P) = \exp\Bigl( {1 \over 2\pi}\int_0^{2\pi} \log|P(e^{it})|dt \Bigr) = |a_n|\prod_{i=1}^n \max(1,|\alpha_i|). \leqno (1) \] If \(|P|_2\) is the \(L^2\)-norm of \(P\) on the circle, then a theorem of Szegö states that \(M(P) = \inf \{|PQ|_2: Q \text{monic}\}\). \textit{W. Lawton} [Proc. Am. Math. Soc. 49, 47-50 (1975; Zbl 0306.12002)] proposed the calculation of \(M(P)\) by computing the approximants \[ M_k(P):= \inf \{|PQ|_2: Q \text{monic, degree}(Q) = k\}, \] which can be computed without computing the roots of \(P\). Clearly \(M(P) \leq M_k(P)\), and \(M_k(P) \to M(P)\) as \(k \to \infty\), but Lawton did not give an estimate of the rate of convergence of \(M_k(P)\) to \(M(P)\). This has now been provided by the author's sharp result: \[ M_k(P) \leq \Bigl( \prod_{i=0}^{n-1} \bigl( 1 + {n \over n - i + k}\bigr)\Bigr)^{1/2} M(P). \leqno (2) \] The proof gives a characterization of the optimal polynomial, which can be used to conveniently compute \(M_k(P)\). However, the rate of convergence given by (2) suggests that this may not be a practical method for the accurate computation of \(M(P)\). It would be interesting to (theoretically and experimentally) compare this method with the two methods suggested by the two formulas in (1).