# Estimation of a sum of algebraic numbers

Let $$\alpha_1, \ldots, \alpha_n$$ be algebraic numbers and let $$p_1, \ldots, p_n$$ be the corresponding minimal polynomials with integer coefficients.
Denote by $$H$$ the maximal magnituge among all coefficients of all $$p_i$$.

Assume that $$\alpha_1 + \ldots + \alpha_n \not= 0$$. Is it true that $$|\alpha_1 + \ldots + \alpha_n | \ge {(CH)}^{-q(\text{deg}(p_1) +\ldots \text{deg}(p_n) )}$$ for some fixed polynomial $$q$$ and constant $$C$$?

Warning: $$C$$ and $$q$$ should not depend on $$\alpha_1, \ldots, \alpha_n$$.

UPD1: it seem that this question is rather difficult. Let us consider a particular case: when all $$\alpha_i$$ are quadratic irrationality. Is it possible to prove the required result about the measure of its linear independence?

UPD2: Is it possible to prove the statement under some reasonable hypothesis like effective abc-conjecture or Vojta-conjecture?

UPD3: It is possible to prove that there exist some $$C$$ such that $$|\alpha_1 + \alpha_2| \ge H^{-C(\deg(p_1) + \deg(p_2))}$$

(if $$\alpha_1 + \alpha_2 \not=0$$)?

In the answer below I assume that $$q$$ may depend on $$n$$. It seems plausible that the assertion is true for $$q(x)=x^2$$ or something similar, but I don't know how to prove it. It may be hard. The answer below is based on Liouville's inequality, that is on estimates for $$P(\alpha_1,\dots,\alpha_n)$$ which are essentially the best possible for arbitrary polynomials and algebraic numbers. But for a fixed $$P$$ much better estimates can be available as illustrated by Roth's theorem. Unfortunately, Roth's theorem and Schmidt's Subspace theorem are ineffective, so can't be used here.

Yes, that's true. I will write the estimates in terms of height. The usual height $$H(p)$$ of a polynomial $$p$$ with integer coefficients is defined as the maximum of absolute values of its coefficients. Then the usual height $$H(\alpha)$$ of an algebraic number $$\alpha$$ is the usual height of its minimal polynomial. The absolute logarithmic height of $$\alpha$$ is $$h(\alpha)=\frac{1}{d}\log H(\alpha)\,,$$ where $$d$$ is the degree of $$\alpha$$.
The basic version of Liouville's inequality says that if $$\alpha\neq0$$ then $$|\alpha|\geq\frac{1}{H(\alpha)}\,.$$ Now, combine that with $$h(\alpha_1+\dots+\alpha_n)\leq\log n+h(\alpha_1)+\dots+h(\alpha_n)$$ and $$\deg(\alpha_1+\dots+\alpha_n)\leq \deg(\alpha_1)\cdot\dots\cdot\deg(\alpha_n)\,.$$
• As I understand your estimate gives something like $H^{-\deg(\alpha_1)\cdot\dots\cdot\deg(\alpha_n)}$. – Alexey Milovanov May 20 '19 at 13:17
• However, I wanted estimate $H^{q (\deg(\alpha_1) + \ldots + \deg(\alpha_n))}$, where $q$ is a fixed polynomial,i.e. $q$ does not depend on $n$ and $\deg (\alpha_i)$. – Alexey Milovanov May 20 '19 at 13:20
• If $q$ can not depend on $n$, then you can find a counterexample already with rational numbers. – Oleg Eroshkin May 20 '19 at 13:28
• So, for every polynomial $p$ and for every $n$ there are $n$ rational numbers $\alpha_1, \ldots, \alpha_n$ such that $|\alpha_1 + \ldots \alpha_n| \ge H^{- p(n)}$, where $H$ is the maximum value among numerators and denominators of $\alpha_i$? – Alexey Milovanov May 20 '19 at 13:39