be a monic polynomial with integer coefficients.
. Then in fact
.
The proof is quite simple using algebra. Suppose
![p(x),q(x) \notin \mathbb{Z}[x]](http://latex.codecogs.com/png.latex?p(x),q(x) \notin \mathbb{Z}[x])
; then there is a least positive integer

and a least positive integer

such that
)
and
)
are in
![\mathbb{Z}[x]](http://latex.codecogs.com/png.latex?\mathbb{Z}[x])
. It's easy to see that there is no integer

which divides all the coefficients of
)
(or else

could be made smaller), and similarily for
)
.
Take any prime

which divides the product

, and let
 \mapsto \overline{h(x)})
denote the canonical projection
![\mathbb{Z}[x] \longrightarrow (\mathbb{Z}/p\mathbb{Z})[x]](http://latex.codecogs.com/png.latex?\mathbb{Z}[x] \longrightarrow (\mathbb{Z}/p\mathbb{Z})[x])
. Then
} = \overline{mnp(x)q(x)} = \overline{mp(x)}\ \overline{nq(x)})
. Since
![(\mathbb{Z}/p\mathbb{Z})[x]](http://latex.codecogs.com/png.latex?(\mathbb{Z}/p\mathbb{Z})[x])
is an integral domain, we must have
} = 0)
or
} = 0)
. But
} \neq 0)
and
} \neq 0)
or else all the coefficients of
)
or all the coefficients of
)
would be divisible by

, which contradicts the choice of

, as described above.