Lemma 4.41.2. (Finite forests outside a semistable vertex set). Let be a split nodal algebraic model with vertex set , skeleton , and retraction . Let be a semistable vertex set, and let be a connected component of meeting . Then is a nonempty finite tree with its boundary endpoints removed. It has one boundary end if is an open ball and two if is an open annulus. Each end has a unique limit in ; the two annular ends are counted separately when their limits coincide. Every special component indexed by is a smooth .
Proof
Proof. We first record the topology of the complement of . Its components are open balls , and For a smooth residual tube this is Proposition 4.22.2. Inside a nodal tube the annular Gauss retraction leaves residue-disc components. In an annular coordinate each such disc is , attached at . Its closure in a compact closed subannulus is the disc together with this Gauss point, by the Gauss seminorm formula and polynomial factorization. This subannulus is compact and hence closed in , so the same is its closure in .
The retraction stays in . The set is open in and closed in , so its boundary lies in . If retracted to , then would lie in a smooth residual -tube attached to : residue discs inside a nodal tube retract to interior edge points, not vertices. That smooth tube is open and closed in and therefore would equal . It contains no vertex of , a contradiction. Consequently avoids . It is connected and contains every , so it lies in . Since fixes , this proves . It is a connected part of a finite graph obtained by deleting vertices in .
There are no circuits. Use an actual ball or annulus coordinate on . Theorem 4.1.1 describes every type II or III point as a Gauss point . At a type II point, removal partitions the analytic projective line into the inward residue discs and the outward disc. To check exhaustiveness, if and , there is with and . Otherwise every linear factor would have its Gauss value at , and factorization over would give . At a type III point, the inward and outward pieces give the analogous two-sided separation. These are disjoint open sets and distinguish the two sides of a radial interval.
The Gauss interval between two points of lies in . In a ball its joining radius is still smaller than the outer radius. In an annulus the value of on the interval is between the two endpoint values, or is their common value. Every interior point of this interval separates its endpoints. Therefore every path between the endpoints contains the interval. An embedded compact arc must equal the interval, since an interval contained in an arc with the same endpoints is the whole arc. This also agrees with the explicit interval calculation of Lemma 4.35.1. A circuit in would give two different arcs between two of its Gauss points. Thus is a tree.
Radial ends have limits in . For a ball , the connected outer tails have nested nonempty compact connected closures in . Their intersection is nonempty and connected. It misses and is contained in its finite boundary , hence is a singleton . Every neighborhood of contains a whole tail: otherwise the nested compact sets outside that neighborhood would have a common point. The closed disc is compact in and closed in . It follows that every boundary point lies in every , so . For an annulus, the same argument applies to each of its two radial ends. A compact middle subannulus shows that its boundary consists of the two resulting limits ; they may be equal.
Each radial end occurs exactly once in the finite tree. Choose a compact disc or middle annulus containing the finite set . Every graph end of leaves all compact subsets of , because its endpoint in the compact graph lies in . Its final portion therefore approaches one coordinate end. The cut-point separation just proved forces this portion to contain a sufficiently short radial Gauss tail. Two different graph ends cannot approach the same coordinate end, since they would share a nonempty Gauss interval.
Suppose a coordinate end were unrepresented. Some sufficiently far tail would then miss . Indeed, if all tails met , one of its finitely many edges would have points escaping toward that end; compactness of the closed edge would produce its endpoint in and a graph end approaching the chosen coordinate end. The connected set lies in one ball component of . Because meets and avoids , we have . Also , so the compact closure misses . All shorter tails lie in , contrary to the nonempty limiting set in proved above. Every coordinate end is therefore represented. This gives exactly one or two ends.
The unwanted components are smooth rational curves. For , translation and scaling identify its type II Gauss field with the unit Gauss field. Its residue field is by Proposition 3.8.1. Proposition 4.24.2 identifies this with the function field of the special component . The normalization is finite by [StacksAlg] (Tag 0BXR). It is therefore proper, and is proper over by [StacksAlg] (Tags 01WM, 01W3). The normal proper curve equivalence [StacksAlg] (Tag 0BY1) now identifies with . A self-node of would give a loop based at in . Its interior is connected, avoids , and meets the open set near , so the entire loop lies in . This contradicts acyclicity. Thus has no self-node. At a node shared with another component, its own branch is smooth, by the pointed branch calculation in Lemma 4.20.2. Hence itself is smooth and equals its normalization, as asserted. ◻