Proposition. (An invariant midpoint over a dense value group). There is a complete nontrivially real-valued field , with algebraically closed residue field of characteristic zero and dense value group , and a split projective semistable model of , for which the following holds. Its geometric skeleton is an interval of length . Adding its midpoint gives a semistable geometric vertex set fixed by every isometric -automorphism of , but no semistable original- model has that geometric vertex set. The midpoint is realized after the ramified quadratic extension .
Proof
Proof. Fix an algebraically closed field of characteristic zero. Take compatible Laurent-series fields and let be the completion of . For nonzero , choose with ; then . Thus completion introduces no new nonzero values and , which is dense in . For , approximation with error less than identifies its residue with a residue from some . Hence the residue field is .
On , with affine coordinate , blow up the ideal , extended as the unit ideal outside its support. The two affine charts over the -line are Their standard monomial bases make these rings free over , hence flat; the displayed presentations are finite. The first is a nodal chart and the second is smooth. Together with the unchanged infinity chart they give a projective semistable model , by Stacks Tag 02NS and Stacks Tag 0C4P, with special fibre two copies of meeting at one split node. Its geometric vertices have , and the node tube is , of length .
Choose with , and put . Since , and . The two terms in , for , have distinct valuations when nonzero, since . Thus ; completeness follows coefficientwise from completeness of . Since , the extension is ramified. Lemma C.1.84, using the accepted charts
Proof input. cand:symmetricnodeblowup
, at on inserts exactly the midpoint. Consequently its base change to realizes the three-element semistable geometric vertex set Each of these Gauss seminorms is fixed by every isometric -automorphism of : its radius is fixed and its formula on uses only the absolute values of the coefficients.
If an original semistable model realized this set, then
Proof input. v036:f5:vertexarithmetic
would make the image of the midpoint a component vertex satisfying . But is a nonzero element of with , a contradiction. Thus invariance, even with already split branches and residue field, is insufficient. The use of above is a ramified construction, not an etale descent step. ◻