NAG-002699 · Accepted whole

A nonflat pointed germ with no finite saturated box

I Descent of the polarized mixed tower

Open in the English reader · PDF page 2111

Statement · exact TeX
\begin{proposition}[A nonflat pointed germ with no finite saturated box]
\label{psc:cutobstruction}
Let \(K=\mathbb Q\) with its trivial absolute value, \(d\geq1\),
and use one puncture block \(0,1\), with \(k=d-1\).
Fix \(\varepsilon>0\) and \(0<r<1\), and define the affinoid domain
\[
 X_r=Y_\varepsilon\cap\{|t-2|\geq r\},\qquad
 f=(t,t-1,z_1,\ldots,z_{d-1}).
\]
Let \(x=\eta\). Then \(X_r\) is pure of dimension \(d\),
\(x\in\operatorname{Int}(X_r/K)\), and the pointed moment
fan at \(x\) is the nonflat three-ray fan times
\(\mathbb R^{d-1}\), with graded rank \(d\).

For every \(a,b>0\), the map
\[
 \psi_{a,b}=(t^a(t-1)^b,z_1,\ldots,z_{d-1})
\]
is finite at \(x\) in the ordinary-neighbourhood sense.
Its full fibre over the target unit Gauss point is \(\{x\}\).
Nevertheless, for \emph{no} full-dimensional closed rectangular
box \(P\) with \(0\in\operatorname{int}(P)\) is the saturated map
\[
     (Q_{a,b}t_f)^{-1}(P)\longrightarrow\mathbb T_P
 \tag{11}\label{psc:impossible}
\]
norm-admissible finite.
This failure already occurs separately for \(Q_{1,1}\) and
\(Q_{1,2}\); hence the same fan, common analytic Gauss set,
interior moment point, and ordinary local finiteness do not
imply the two finite-collar hypotheses.
\end{proposition}
Full selected proof · exact TeX
\begin{proof}
The lower bound on \(t-2\) is a Laurent-domain condition.
Domain purity gives the asserted dimension.
At \(x\) its inequality is strict, since \(|t-2|(x)=1>r\);
all original moment bounds are also strict.
Adjoining the inverse of \(t-2\) with radius \(r^{-1}>1\)
to the presentation in Lemma~\ref{psc:basecollars} witnesses
absolute interiority of \(x\).
As \(X_r\) is an ordinary neighbourhood of \(x\) in
\(Y_\varepsilon\), full pointed stabilization and
Theorem~\ref{psc:simultaneous} with the identity cover give
the same rank and fan. The fibre assertion follows from
(3), since \(x\) remains in the domain.

For completeness, ordinary local finiteness can be exhibited
using only the frozen local ingredients.
Choose a small box \(P_0\) satisfying (2), and let
\(Y_0=q_{a,b}^{-1}(\mathbb T_{P_0})\subset Y_\varepsilon\).
It is compact and finite over the full polyannulus.
Inside \(Y_0\), the open set \(G=\{|t-2|>r\}\) contains its
\emph{entire} central analytic Gauss fibre \(\{x\}\).
The image of \(Y_0\setminus G\) is compact, hence closed in
the Hausdorff target, and misses the target Gauss point.
Choose an ordinary affinoid target neighbourhood \(V\) of that
point, inside the complementary open set and the ordinary
interior of \(\mathbb T_{P_0}\).
The ordinary affinoid-neighbourhood basis is the one used in
\Dep{v037:forms:germsheaf}.
Finite base change gives \(q_{a,b}^{-1}(V)\to V\) finite,
and its full inverse image lies in \(G\subset X_r\).
It is an ordinary neighbourhood of \(x\).
The domain-interior and composition formulae (with \hyperref[fnd:finiteflatordinaryopen]{ordinary openness} for this finite free map) identify the
ordinary interior of \(q_{a,b}^{-1}(V)\) in \(Y_0\) with the
inverse image of the ordinary interior of \(V\) in
\(\mathbb T_{P_0}\): compute relative interior over
\(\mathbb T_{P_0}\) using the two finite factorizations.
Since \(V\) lies in the ordinary interior of that polyannulus,
these are also ordinary open neighbourhoods in the ambient
source and target. Restricting gives a finite map between them,
as required by finiteness at \(x\).
Only this ordinary target neighbourhood, not a tropical box,
has been constructed.

Suppose (11) were norm-admissible finite.
Restrict its target to a sufficiently small closed box about
zero. Finite base change preserves the supposed property.
We may therefore assume that \(P\) satisfies (2).
Set
\[
 Y_P=q_{a,b}^{-1}(\mathbb T_P),\qquad
 C=X_r\cap Y_P.
\]
Then \(C\) is an affinoid domain of \(Y_P\), and is the
entire inverse image in (11).

For \(0<\rho<1\), let \(y_\rho\) be the simultaneous Gauss
point in the coordinates
\((t-2,z_1,\ldots,z_{d-1})\), of radii
\((\rho,1,\ldots,1)\).
The trivial ground norm gives
\[
 |t(y_\rho)|=|t(y_\rho)-1|=|z_b(y_\rho)|=1,\qquad
                         |t(y_\rho)-2|=\rho.
\]
Every \(y_\rho\) lies in \(Y_P\) and has original and
projected weight zero. The map \(\rho\mapsto y_\rho\) is
continuous: polynomial Gauss norms are continuous maxima,
the puncture inverses have norm \(1\), and uniform
approximation in an ambient affinoid presentation handles
every analytic function.
But \(y_\rho\in C\) exactly when \(\rho\geq r\).
Thus \(y_r\in C\) is not in the ordinary interior of \(C\)
inside \(Y_P\).

The supposed norm-admissible finiteness of \(C\to\mathbb T_P\)
makes every point relatively interior, by
\Dep{fnd:generalizedfinitealgebras}.
Apply the frozen composition identity (using the \hyperref[fnd:nag001626:source-factor-interior]{source-factor implication}) to
\(C\hookrightarrow Y_P\to\mathbb T_P\):
\[
 \operatorname{Int}(C/\mathbb T_P)
 =\operatorname{Int}(C/Y_P)\cap
       (C\hookrightarrow Y_P)^{-1}\operatorname{Int}(Y_P/\mathbb T_P).
\]
It forces \(y_r\in\operatorname{Int}(C/Y_P)\).
The domain-interior theorem
\Dep{fnd:affinoiddomainordinaryinterior} says this is exactly
ordinary interior, contradicting the preceding path.
This proves the obstruction.
\end{proof}
Typeset reader excerpt
Typeset mathematical passage from PDF page 2111
PDF page 2111 · excerpt continues in the reader

Typed proof inputs

Reader cross-references

Proof map

Verified map inputs · NAG-002699

AND · joint registered proof inputs