Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Schinzel nd two theorems gelfond applications
Schinzel, A., On two theorems of Gelfond and some of their applications. Acta Arith. 13 (1967/68), 177-236. No notice is printed in the file; the publisher's article record offers the PDF "Free download under CC-BY license" ("Pobierz zgodnie z CC-BY" as the Polish page prints it), no version named (https://www.impan.pl/get/doi/10.4064/aa-13-2-177-236, read 2026-10-02): the Creative Commons Attribution license with no version named.
Schinzel reworks Gelfond's ordinary and p-adic measures of irrationality for the ratio of two logarithms of algebraic numbers so that the resulting bounds are explicit rather than depending on unspecified functions n_0(eps, alpha, beta). Section 2 reproduces Gelfond's 1940 arguments with modifications replacing log^{3+eps} max{|n|,|m|} by C(alpha,beta)(log max{|n|,|m|} + C'(alpha,beta))^3, or its p-adic analogue, in the inequalities for G_0 and G_p; Theorems 1 (p-adic) and 2 (ordinary) write these constants out explicitly, and when log|alpha|/log|beta| is rational he obtains G_0 > -C(alpha,beta)(log max{|n|,|m|} + C'(alpha,beta))^2, the paper's one improvement on Gelfond's 1952 work, reformulated as Corollary 1 in Diophantine-approximation terms. Section 3 applies this to second-order linear recurrences: Theorems 3 and 4 give explicit lower estimates for |u_n| when the companion polynomial has non-real roots, comprising earlier results of P. Chowla, S. Chowla, Dunton, Lewis, Townes and Schinzel. Theorem 5 shows that for a negative odd integer d other than 1 - 2^k the equation x^2 - d = 2^m has at most one solution with m > 80 and x > 0, so that the Browkin-Schinzel conjecture (at most one solution in positive integers for d other than 1 - 2^k and -23) reduces to a finite computation; Theorem 6 proves the analogue for x^2 - d = p^m with p a prime factor of 1 - 4d, and Theorems 7 and 8 estimate the greatest prime factor of u_n. Section 4 treats x^nu - eps P_1^{n_1} ... P_k^{n_k} with nu = 2 or 3, eps = ±1 and P_i positive integers: Theorem 9 bounds its absolute value from below, Theorem 10 bounds its greatest prime factor from below for k <= 3 under stated restrictions, and Corollary 6 uses Theorem 10 to solve effectively Diophantine equations of the form q_1^{y_1} ... q_i^{y_i} ± r_1^{z_1} ... r_j^{z_j} = s^x, with the q's and r's distinct primes and s >= 1, where 6 does not divide s if the sign is minus. Corollary 5 to Theorem 9 gives, for a real quadratic irrational xi and an integer g > 1, the effective bound ||xi g^n|| > g^{-n} exp(c n^{1/7}), slightly more than Liouville's theorem yields. Section 5, "The greatest prime factor of a quadratic or cubic polynomial", bounds q(Ax^nu - E) from below by a multiple of log log x for nu = 2, 3 (Theorem 11, with Corollary 7 for any quadratic polynomial without a double root), improves the constants of Mahler and Nagell for Ax^2 - E with E | 4 (Theorem 12), and in Theorems 13-15 goes the opposite way, showing that for suitable x the greatest prime factor of Ax^nu - E, and more weakly of any integer polynomial f(x) of degree greater than 1, is small compared with the value; this is the aspect bearing on problems #368 and #928, which cite it as [Sc67b]. The paper closes with an open problem, and Schinzel notes that Baker's 1966 solution of the three-logarithms problem would generalize many of these results.
Source: https://doi.org/10.4064/aa-13-2-177-236.
Results to transcribe.
- Theorems 1 and 2: Explicit versions of Gelfond's p-adic (Theorem 1) and ordinary (Theorem 2) irrationality measures for the ratio of two logarithms, with log^{3+eps} N, N = max{|n|,|m|}, replaced by an explicit constant times (log N + C')^3; in Theorem 2 the exponent 3 becomes 2 when |alpha| and |beta| are multiplicatively dependent (when alpha and beta are, if |alpha| = |beta| = 1).
- Corollary 1: For an algebraic integer gamma != 0 with gamma/|gamma| not a root of unity and H = max{|n_1|,|n_2|} > 1, |arg(gamma)/(2 pi) - n_1/n_2| > exp(-c(gamma) log^2 H), with c(gamma) independent of n_1, n_2.
- Theorems 3 and 4: Explicit lower estimates for |u_n| for second-order linear recurrences with non-real companion-polynomial roots, comprising earlier estimates of Chowla, Dunton, Lewis, Townes and Schinzel.
- Theorem 5: For a negative odd integer d != 1 - 2^k, the equation x^2 - d = 2^m has at most one solution with m > 80, x > 0.
- Theorem 10: Under each of four stated conditions, all with k <= 3, a lower bound for the greatest prime factor of x^nu - eps P_1^{n_1} ... P_k^{n_k}; Corollary 6 derives from it the effective solution of Diophantine equations q_1^{y_1} ... q_i^{y_i} ± r_1^{z_1} ... r_j^{z_j} = s^x (q's and r's distinct primes, s not divisible by 6 when the sign is minus).
- Theorems 11 and 12, Corollary 7 and Theorems 13-15: Lower bounds of order log log x for the greatest prime factor of quadratic and cubic binomials Ax^nu - E (Theorem 11) and of quadratic polynomials (Corollary 7), Theorem 12 improving Mahler's and Nagell's constants for Ax^2 - E with E | 4, together with upper bounds showing how small the greatest prime factor of a binomial (Theorems 13 and 14) or of a general integer polynomial (Theorem 15) can be for suitable x.