For a given polynomial $f(x_1, \cdots, x_n) \in \mathbb{Z}[x_1, \cdots, x_n]$, define the height of $f$ as $H(f)$ as the maximum absolute value among its coefficients. We can also define the log height $h(f) = \log H(f)$.

If $f = f_1 \cdots f_r, f_j \in \mathbb{Z}[x_1, \cdots, x_n]$ and $d = \deg f$, then Gelfond's inequality states that $$\displaystyle \sum_{j=1}^r h(f_j) \leq h(f) + nd.$$

My question is the following: Suppose $F(x_0, \cdots, x_n) \in \mathbb{Z}[x_0, \cdots, x_n]$ is homogeneous with degree $d$ and is primitive (i.e., its coefficients are co-prime), which is irreducible over $\mathbb{Q}$ but is not absolutely irreducible. Let $K/\mathbb{Q}$ be a number field such that $F$ splits completely into absolutely irreducible factors $F_1, \cdots, F_r$. Then the $F_i$'s are necessarily conjugate. Further, for each $j, 1 \leq j \leq r$ we can write $$\displaystyle F_j = \sum \lambda_{i,j} F_{i,j}$$ where the $\lambda_{i,j}$'s are $\mathbb{Q}$-linearly independent and the $F_{i,j} \in \mathbb{Z}[x_0, \cdots, x_n]$ for all $i$. Indeed, by noting that the $F_j$'s are conjugate we see that the $F_{i,j}$'s don't depend on $j$. Let $G = F_{u,v}$ be such that $G$ has the smallest (projective) height among all $F_{i,j}$'s. Can we bound $H(G)$ or $h(G)$ in terms of $F$?


I think the answer is no, at least in the currently stated form. For instance, consider $x^2 + y^2 = (x+ iy)(x-iy)$. Now choose a matrix in $SL_2(\bf Z)$, i.e. choose $a,b,c,d$ integers such that $ad - bc = 1$. Let $x' = ax + by$ and $y' = cx + dy$ be linear forms. Then $x = dx' - by'$ and $y = -cx' + ay'$. So we get $$ x^2 + y^2 = \big( dx' - by' + i(-cx' + ay') \big) \big( dx' - by' - i(-cx' + ay') \big) = \big( (d - ic)x'+ (a-ib) y' \big) \big( (d+ic) x' - (a+ib) y' \big) $$ And now $x'$ and $y'$ have large height, if $a,b,c,d$ are large enough. So unless you have a better way of pinning down (a choice of) $F_{i,j}$, I don't think you can get an upper bound on their heights. (Maybe there is a bound if you take the minimum over all choices?)

| cite | improve this answer | |
  • $\begingroup$ I believe you are right. I think the question is more reasonable if one supposes some sort of structure on the $F_{i,j}$'s. One way to think about it is to consider an integral basis $c_1, ..., c_d$ of the ring of integers in $\mathcal{O}_K$, then take a dual basis $\omega_1, \cdots, \omega_d$. Then one can write $F_j = \sum_{i=1}^d \lambda_i \operatorname{Tr}(F_j \omega_i)$. One can hopefully get a bound in terms of $F_j$ and the basis. $\endgroup$ – Stanley Yao Xiao Mar 16 '15 at 16:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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