5
$\begingroup$

Consider a degree-$d$ algebraic integer $\alpha$ all of whose conjugates (including itself) are real numbers greater than 1. Its Mahler measure $M(\alpha)$ is simply equal to the norm $N(\alpha)$. Since $N(\alpha-1) \geq 1$, an application of $H\ddot{o}lder's$ inequality gives the lower bound $N(\alpha) \geq 2^d$, with equality achieved only by $\alpha = 2$.

Question. Does there exist $C > 2$ such that $N(\alpha) \geq C^d$ for all $\alpha \neq 2$?

$\endgroup$
7
  • $\begingroup$ Interesting question. I assume you're familiar with work of Smyth, Flammang, etc. on lower bounds for the Mahler measure of totally positive algebraic integers? (These give a lower bound of $C=1.722...$ for $M(\alpha)^{1/d}$ with finitely many exceptions, but valid for all totally positive integers $\alpha$, not just those with all conjugates $>1$). $\endgroup$ Apr 20, 2016 at 21:13
  • $\begingroup$ On the hypothesis that $\alpha$ is algebraic of degree $d$, the value $\alpha=2$ can only arise in the case $d=1$, so the equality $N(\alpha)=2^d$ is achieved only for $d=1$, $\alpha=2$. $\endgroup$ Apr 20, 2016 at 22:58
  • $\begingroup$ @BobbyGrizzard Is the "with finitely many exceptions" part effective? $\endgroup$
    – Fan Zheng
    Apr 21, 2016 at 5:50
  • $\begingroup$ @FanZheng yes. Here is one effective result: ams.org/journals/mcom/1996-65-213/S0025-5718-96-00664-3/… edit: and here is the one I was probably looking at yesterday, due to Wu and Mu (Quanwu Mu, not the OP!): sciencedirect.com/science/article/pii/S0022314X12001989 $\endgroup$ Apr 21, 2016 at 11:58
  • $\begingroup$ @BobbyGrizzard NOT the OP LOL $\endgroup$
    – Fan Zheng
    Apr 22, 2016 at 4:19

2 Answers 2

5
$\begingroup$

The answer is yes.

Suppose that $\alpha$ is totally real algebraic integer, and that all its conjugates are greater than $1$. Suppose also that $\alpha \ne 2$.

Note the elementary inequality for $x \in (1,\infty) \setminus \{2\}$:

$$\ln|x| \ge \frac{\ln|x-1| + \ln|x-2|}{3} + \ln C,$$

where $C = 2^{1/3} 3^{1/2} = 2.18 \ldots > 2$. (Equality holds precisely at $3 \pm \sqrt{3}$.)

Denote the conjugates of $\alpha$ by $\alpha_i$. Assuming that $\alpha \ne 1,2$, we see that, because $\alpha - 1$ and $\alpha - 2$ are non-zero algebraic integers, we have inequalities

$$\sum_{i=1}^{d} \ln | \alpha_i - 1| = \ln N(\alpha -1) \ge 0,$$ $$\sum_{i=1}^{d} \ln |\alpha_i - 2| = \ln N(\alpha -2) \ge 0.$$

Hence we deduce that $\displaystyle{\ln N(\alpha) = \sum_{i=1}^{d} \ln |\alpha_i| \ge d \cdot \ln C}$, and thus $N(\alpha) \ge C^{d}$ where $C > 2$.

$\endgroup$
1
  • $\begingroup$ I've seen a one page proof of Lehmer's conjecture for totally real by Hohn and Skoruppa using such type of inequality. How does one come up with such "elementary inequality"? $\endgroup$
    – Kalas678
    Dec 26, 2023 at 19:32
3
$\begingroup$

Gypsum's argument is really nice. In the same spirit, we have the following inequality: $$\ln|x| \geq \frac{2\ln|x-1| + \ln|x-2|}{5} + \ln\sqrt{5}$$ which is achieved at $x = \frac{5\pm \sqrt{5}}{2}$. This way we obtain the optimal constant $C = \sqrt{5}$. Perhaps some further tweaks can yield even larger $C$ with finitely many exceptions, as in the work of Smyth and Flammang.

I wonder how far this approach can be extended. Smyth showed that for totally positive algebraic integers (whose conjugates are not necessarily greater than 1), their $M(\alpha)^{\frac{1}{d}}$ are dense beyond 1.73, which is very close to the lower bound cited by @BobbyGrizzard. Do we have a similar situation here? As a first step, it would be good to construct infinitely many $\alpha$ that give upper bound to $C$. For example, consider the $n$-th $Chebyshev$ polynomial $T_n(x)$. Then the monic polynomial $(-x)^nT_n(\frac{2}{x}-1)$ has all its roots greater than 1. In this case $N(\alpha) = 2^{2n-1}$, suggesting $C \leq 4$.

$\endgroup$
3
  • $\begingroup$ Welcome to mathoverflow! This is really a great question, but as you are new here, please be informed that the answers are just "answers"; new questions included in answers may not get enough attention. If you want to probe further, please post a separate question; you may include a link back to this question. $\endgroup$
    – Fan Zheng
    Apr 22, 2016 at 6:25
  • $\begingroup$ One can usually extend this approach a certain amount, but not as far as the limit. The fact that equality is achieved at $x = (5+\sqrt{5})/2$ with the optimal bound is related to the fact that $x - 1$ and $x - 2$ are both units. But there is no reason this had to happen. BTW, to improve the bound beyond this point (with the exceptions $x = 2$ and $(5 \pm \sqrt{5})/2$, you simply need to add a very small multiple of the term $\log(x^2 - 5 x + 5)$. $\endgroup$
    – Gypsum
    Apr 22, 2016 at 13:11
  • $\begingroup$ @FanZheng Thanks for your advice :) $\endgroup$ Apr 22, 2016 at 17:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.