TITLE: How small can a totally positive integer be?
QUESTION [9 upvotes]: Consider a large, fixed $M>2$. For each $n$, let $\alpha_n$ denote the smallest algebraic integer of degree at most $n$, all of whose Galois conjugates lie in the real interval $(0,M)$. 
Is there anything known on the rate at which $\alpha_n$ decays to $0$ as $n\to\infty$ ? 
The exponential lower-bound $\alpha_n>\frac1 {M^{n}}$ is not hard to prove, but is it reasonable to expect that $\alpha_n$ will actually decay much more slowly, i.e. like an inverse power of $n$ ?

REPLY [12 votes]: $\alpha_n$ can be exponentially small once $M$ is large enough,
say $M \geq 6$.
For $m > 0$ let $\tau_m$ be the (monic, degree-$m$) polynomial
such that $\tau_m(z+z^{-1}) = z^m + z^{-m}$; in other words,
$\tau_m(x) = 2 T_m(x/2)$ where $T_m$ is the $m$-th Čebyšev polynomial.
Then $\tau_m$ has all its roots real and contained in the interval $(-2,2)$,
while there are $m+1$ points $x_0,x_1,\ldots,x_m$ with
$$
2 = x_0 > x_1 > x_2 > \ldots > x_m = -2
$$
at which $\tau_m(x_k) = (-1)^k \cdot 2$.
Now fix an integer $M>5$, for $n$ odd let $m$ be the even number $n-1$
and consider $P(x) = x \, \tau_m (x-(M-2)) - 1$.
This is a monic polynomial with $P(1)=-1$, $P(M-4)>0$, and
$P$ changing sign between $x_{k-1}$ and $x_k$ for each
$k=1,2,\ldots,m$.  Therefore it has $m+1=n$ real roots,
all in $(0,M)$, and all but one greater than $M-4$.
Moreover the product of these roots is $-P(0) = 1$.
Hence the remaining root is smaller than $(M-4)^{1-n}$,
which decays exponentially once $M-4>1$.  (In fact $M=5$
is good enough because about half the roots exceed $M-2$.)
For example, if $M=10$ and $n=7$ this recipe yields the septic polynomial
$$
x^7 - 48x^6 + 954x^5 - 10048x^4 + 59145x^3 - 184464x^2 + 238142x - 1,
$$
with one root just below $4.2 \cdot 10^{-6}$
and the other six ranging from about $6.061$ to $9.936$.