NAG-003037 · Accepted whole

Normalization preserves the coefficient with branch weights

Proper direct image on historical ordinary compact tests

Open in the English reader · PDF page 1810

Statement and full selected proof · exact TeX
\begin{lemma}[Normalization preserves the coefficient with branch weights]
\label{f9cn:branches}
Let $A$ be a one-dimensional Noetherian local domain whose integral
closure $B$ in $F=\Frac(A)$ is finite over $A$. Let
$\mathfrak q_1,\ldots,\mathfrak q_r$ be the maximal ideals of $B$.
Their contractions are the maximal ideal $\mathfrak m$ of $A$;
each $B_{\mathfrak q_j}$ is a DVR. For every $f\in F^\times$,
\begin{equation}\label{eq:branches}
 \ord_A(f)=\sum_{j=1}^r
     [\kappa(\mathfrak q_j):\kappa(\mathfrak m)]
                  \ord_{B_{\mathfrak q_j}}(f).
\end{equation}
In particular, for nonzero $t\in\mathfrak m$,
\begin{equation}\label{eq:f9cn:branch-length}
 \len_A(A/tA)=\sum_j[\kappa(\mathfrak q_j):\kappa(\mathfrak m)]
                          \len_{B_{\mathfrak q_j}}(B_{\mathfrak q_j}/t).
\end{equation}
All residue degrees are full degrees; no separable-degree substitution
or reduction of the quotient rings is permitted.
\end{lemma}
\dep{Finite module length, the affine integral-closure construction,
and the proposed \st{00PD}; no determinant or pending F9 claim.}
\begin{proof}
The finite algebra $B/\mathfrak mB$ is Artinian. A prime of $B$
over $(0)$ is $(0)$, since localization by $A\setminus\{0\}$
gives the field $F$. A prime over $\mathfrak m$ is maximal,
because its quotient is a finite domain over $A/\mathfrak m$, hence
a field. A maximal ideal cannot lie over $(0)$: this would make $B$
a field finite over $A$; a monic equation for $1/t$, with
$0\ne t\in\mathfrak m$, would then imply $1/t\in A$.
Thus $B$ is Noetherian and semilocal, and each of its maximal
localizations is a normal local domain of dimension one.
Tag~\href{https://stacks.math.columbia.edu/tag/00PD}{00PD}
gives the DVR assertion. The residue extensions are finite because
$B$ is finite over $A$.

The finite $A$-module $Q=B/A$ is torsion and has finite length.
Indeed, clearing the denominators of finitely many generators kills
$Q$ by a nonzero element of $A$; in dimension one its support is
contained in $\{\mathfrak m\}$. Multiplication by $t$ is injective
on both $A$ and $B$. The kernel-cokernel sequence gives
\[
 0\longrightarrow Q[t]\longrightarrow A/tA\longrightarrow B/tB
       \longrightarrow Q/tQ\longrightarrow0.
\]
On a finite-length module the kernel and cokernel of any endomorphism
have the same length. Hence $\len_A(A/tA)=\len_A(B/tB)$.
The finite-length $B$-module $B/tB$ has a composition series with
simple factors $\kappa(\mathfrak q_j)$, their numbers being the
lengths after localization at $\mathfrak q_j$. The $A$-length of
such a factor is its full dimension over $\kappa(\mathfrak m)$.
This proves \eqref{eq:f9cn:branch-length}. The same equality for a nonzero unit
has both sides zero. Apply it to $a$ and $b$ and subtract to obtain
\eqref{eq:branches}.
\end{proof}
Typeset reader excerpt
Typeset mathematical passage from PDF page 1811
PDF page 1811

Typed proof inputs

Proof map

Verified map inputs · NAG-003037

No addressable selected proof input is recorded in this projection.

Standing context and base inputs
  • BASE-RINGS