Broccoli curves and the tropical invariance of Welschinger numbers (Q390747)

This paper continues the development of tropical methods to determine certain enumerative invariants. Prior to tropicalization, the objects one is interested in are real rational curves in a given ample linear system on a real toric del Pezzo surface passing through a conjugate-invariant collection of points of the appropriate cardinality. The number of such curves is a so-called Welschinger invariant, a sort of \(\mathbb{R}\)-analogue of Gromov-Witten invariants, and it is known that these are indeed invariant: they depend only on the number of real points, not their locations. \textit{G. Mikhalkin} [J. Am. Math. Soc. 18, No. 2, 313--377 (2005; Zbl 1092.14068)] (in the case all points are real) and \textit{E. Shustin} [J. Algebr. Geom. 15, No. 2, 285--322 (2006; Zbl 1118.14059)] (generalizing to the conjugate-invariant collections considered here) proved a ``Correspondence Theorem'' that these invariants can be viewed instead as counting certain tropical curves (metric graphs), with appropriate multiplicity, in the corresponding tropical toric surface. As with the celebrated Correspondence Theorem in the complex case, this allows for purely combinatorial methods, leaving the realm of algebraic geometry, to compute a number that on the face of it is a purely algebro-geometric quantity. Needless to say, however, these combinatorial methods can become complicated quite quickly, and they lack the aesthetic elegance of a proper intersection theory and quantum cohomology perspective as on the pre-tropicalized side of the equation. This interpretation of this tropical count via the Correspondence Theorems as an enumerative \textit{invariant} implies that the number of tropical curves one counts is indeed invariant of the location of the points, a fact which is not at all obvious directly from the tropical setup and definitions. More precisely, in the case all points are real (which, incidentally, means that the tropical curves one counts are the same as the ones counted to compute GW invariants, only their multiplicities are modified) there is a simple, direct argument for the desired invariance; in the more general conjugate-invariant case, however, the situation is significantly more complicated and is the main subject of the present paper. One advantage of this, and result in the paper, is a Caporaso-Harris style recursive formula. Another interesting issue that arises is that a natural ``relative'' version of these counts is \textit{not} invariant -- but by mimicking ideas in Caporaso-Harris' original paper [\textit{L. Caporaso} and \textit{J. Harris}, Invent. Math. 131, No. 2, 345--392 (1998; Zbl 0934.14040)] there is a distinguished collection of points, far from general position, which yield an interesting and useful curve count, and which is adapted to the tropical setting here. The main idea in the paper is to replace the count of all tropical curves required by the Correspondence Theorem with a special class of curves, humorously termed here broccoli curves. There is not a bijection between the two types of curves, yet by a so-called bridging algorithm developed herein the authors are able to translate between the two types of curves and thereby reduce to the combinatorially simpler, and more manifestly invariant, broccoli curves. Curiously, there is not a known class of real curves on the pre-tropicalized side of the story corresponding to these curves---whether one exists and exhibits interesting properties, or whether these are purely intrinsic, combinatorial objects, is an interesting question to explore in future work.