Riemann-Roch theory for weighted graphs and tropical curves (Q390718)

This paper continues the study of linear systems on nodal limits of smooth curves via the combinatorial formalism of divisor classes and linear equivalence on decorated graphs. Perhaps the most useful review I can provide here is to clarify the role played by the handful of papers in the subject that all bear unfortunately similar titles, thereby emphasizing the advancements that this paper makes. Chip firing games, the discrete Laplacian, and related concepts have been studied by discrete geometers for many years; it was, however, not until the seminal paper [Adv. Math. 215, No. 2, 766--788 (2007; Zbl 1124.05049)] by \textit{M. Baker} and \textit{S. Norine} that the striking connection to Riemann surfaces (algebraic curves) was firmly established. Indeed, those authors considered loop-less graphs, thought of as the dual graph of a Deligne-Mumford stable curve with smooth components, defined a divisor as an integer-linear combination of the vertices, defined linear equivalence in a combinatorial way that can be interpreted as a chip-firing game, proved a Riemann-Roch theorem in this setting, and then went on to study the analogue of the Jacobian and various maps to it, extending classical Riemann surface results to purely combinatorial results relying only on graph theory. Indeed, the methods to prove this Riemann-Roch were purely combinatorial, and at key steps involving ideas that do not appear to have classical algebro-geometric interpretations. This paper opened the flood-gates of research in this area. \textit{A. Gathmann} and \textit{M. Kerber} [Math. Z. 259, No. 1, 217--230 (2008; Zbl 1187.14066)] showed how to extend Baker-Norine's formalism and Riemann-Roch result to a setting where the edges of the graphs are assigned positive real numbers, essentially by first viewing Baker-Norine as the case where all edge-lengths are one, observing that this implies the case where all edge-lengths are rational numbers, and then deducing the final case by perturbing a graph with arbitrary lengths to this rational setting without changing the ranks and other parameters involved. Roughly contemporaneously, \textit{G. Mikhalkin} and \textit{I. Zharkov} [Contemp. Math. 465, 203--230 (2008; Zbl 1152.14028)] stated and proved this same theorem, though with a different proof technique. After this development, further refinements and connections to other fields were discovered. For instance, \textit{J. Hladký, D. Kràl'} and \textit{S. Norine} [``Rank of divisors on tropical curves'', \url{arxiv:0709.4485}] pursued another combinatorial approach to Baker-Norine's Riemann-Roch and this edge-weighted (so-called ``tropical'') variant, whereas \textit{O. Amini} and \textit{M. Manjunath} [Electron. J. Comb. 17, No. 1, Research Paper R124, 50 p. (2010; Zbl 1277.05105)] translated a key combinatorial step of Baker-Norine into the language of lattice ideals and explored it in that setting, then \textit{M. Manjunath} and \textit{B. Sturmfels} [J. Algebr. Comb. 37, No. 4, 737--756 (2013; Zbl 1272.13017)] extended this approach by recognizing the lattice ideals perspective as a special case of the methods developed in combinatorial commutative algebra (see the book of Miller-Sturmfels) and thus reproved the graph-theoretic Riemann-Roch result by viewing it as a duality result in the theory of free-resolutions of binomial ideals. All the while, as these various reproofs of and slight extensions of Baker-Norine's original result were being produced, another seminal step in the theory was proffered by Baker himself: in ``Specializations of linear systems from curves to graphs'' he shows that not only are these graph-theoretic results beautiful analogies of classical algebro-geometric results, but in fact there is a specialization map that takes a divisor class on a smooth curve, follows its limit to a degenerate nodal curve, then yields a divisor class on the associated dual graph; this process provides a concrete bridge between the world of divisors on Riemann surfaces and on graphs. This sheds new light on these Riemann-Roch results, though to date there is not a proof of the classical Riemann-Roch using these methods. The main result here, instead, is that the rank of a linear system can only increase upon specializing to the combinatorial model of a central fiber. One should view this, philosophically, as a kind of upper-semicontinuity, and indeed the proof comes by way of tracking intersection numbers upon specialization and is actually not a difficult result. Some mystery, remains, however, as Riemann-Roch says that the amount the rank of a complete linear system \(|D|\) can increase is exactly the same as the amount that the rank of \(|K-D|\) increases, and this is not at all clear from the specialization result alone. With this specialization interpretation of divisors on graphs in mind, and in particular by viewing edge-weighted graphs as ``tropical curves'' obtained as another kind of semi-algebraic degeneration, one is naturally led to consider graphs with loops and in fact with contracted ``hidden'' loops. Indeed, dual graphs of arbitrary DM-stable curves can have loops, and by decorating edges with weights we have a moduli space of tropical curves, but to compactify it one needs to allow the lengths of the loops and other edges to approach zero, the result being combinatorially encoded as an edge-weighted graph with additional integer weights on the vertices indicating these hidden loops, or put another way, encoding the geometric genera of the irreducible components on the algebraic curve with this graph as its dual graph. This setting, although a very slight extension of the previous ones, required some additional techniques to handle properly, and that is precisely the role of the present paper by Amini-Caporaso. One might make the argument, in fact, that the article is slightly mis-named, as the extension of Riemann-Roch to this vertex-weighted setting really is an almost trivial modification of the previously published cases, whereas it is the extension of Baker's specialization lemma to this setting that appears to be the main non-trivial and technical contribution of this paper. We now better understand the relationship between algebraic curves and their tropical curves, and especially their limits, thanks to the paper of \textit{D. Abramovich, L. Caporaso} and \textit{S. Payne} [Ann. Sci. Éc. Norm. Supér. (4) 48, No. 4, 765--809 (2015; Zbl 1410.14049)], so the current paper of Amini-Caporaso plays an important role in extending the Baker theory accordingly.