## Spec is representable

Saw Matt Baker at the Joint Meetings and he told me about this crazy paper he just posted, “Matroids over Hyperfields.”   A hyperring is just like a ring except addition is multivalued; given elements x and y of R, x+y is a subset of R which you can think of as “the possible outcomes of summing x and y.”  A hyperfield is a hyperring in which every nonzero element has a multiplicative inverse.

Here’s an example familiar to tropical geometers:  let T be the hyperfield whose elements are $\mathbf{R} \bigcup -\infty$, whose multiplication law is real addition, and whose addition law is

a + b = max(a,b) if a <> b

a + b = {c: c < a} if a=b

In other words, each element of T can be thought of as the valuation of an otherwise unspecified element of a field with a non-archimedean valuation, and then the addition law answers the question “what is ord(x+y) if ord(x) = a and ord(y) = b”?

This may sounds at first like an almost aggressively useless generalization, but no!  The main point of Matt’s paper is that it makes sense to talk about a matroid with coefficients in a hyperfield, and that lots of well-studied flavors of matroids can be written as “matroids over F” for a suitable hyperfield F; in this way, a lot of different stories about different matroid theories get unified and cleaned up.

In fact, a matroid itself turns out to be the same thing as a matroid over K, where K is the Krasner hyperfield:  just two elements 0 and 1, with the multiplication law you expect, and addition given by

0 + 0 = 0

0 + 1 = 1

1 + 1 = {0,1}

One thing I like about K is that it repairs the problem (if you see it as a problem) that the category of fields has no terminal object.  K is terminal in the category of hyperfields; any hyperfield (and in particular any field) has a unique map to K which sends 0 to 0 and everything else to 1.

More generally, as Matt observes, if R is a commutative ring, a homomorphism f from R to K is nothing other than a prime ideal of R — namely, f^{-1}(0).  So once you relax a little and accept the category of hyperfield, the functor Spec: Rings -> Sets is representable!  I enjoy that.

Update:  David Goss points out that this observation about Spec and the Krasner hyperfield is due to Connes and Consani in “The hyperring of adèle classes” JNT 131, (2011) 159-194, p.161.  In fact, for any scheme X of finite type over Z, the underlying Zariski set of X is naturally identified with Hom(Spec(K),X); so Spec(K) functions as a kind of generic point that’s agnostic to characteristic.

## What I learned at the Joint Math Meetings

Another Joint Meetings in the books!  My first time in San Antonio, until last weekend the largest US city I’d never been to.  (Next up:  Jacksonville.)  A few highlights:

• Ngoc Tran, a postdoc at Austin, talked about zeroes of random tropical polynomials.  She’s proved that a random univariate tropical polynomial of degree n has about c log n roots; this is the tropical version of an old theorem of Kac, which says that a random real polynomial of degree n has about c log n real roots.  She raised interesting further questions, like:  what does the zero locus of a random tropical polynomial in more variables look like?  I wonder:  does it look anything like the zero set of a random band-limited function on the sphere, as discussed by Sarnak and Wigman?  If you take a random tropical polynomial in two variables, its zero set partitions the plane into polygons, which gives you a graph by adjacency:  what kind of random graph is this?
• Speaking of random graphs, have you heard the good news about L^p graphons?  I missed the “limits of discrete structures” special session which had tons of talks about this, but I ran into the always awesome Henry Cohn, who gave me the 15-minute version.  Here’s the basic idea.  Large dense graphs can be modeled by graphons; you take a symmetric function W from [0,1]^2 to [0,1], and then your procedure for generating a random graph goes like this. Sample n points x_1,…x_n uniformly from [0,1] — these are your vertices.  Now put an edge between x_i and x_j with probability W(x_i,x_j) = W(x_j,x_i).  So if W is constant with value p, you get your usual Erdös-Renyi graphs, but if W varies some, you can get variants of E-R, like the much-beloved stochastic blockmodel graphs, that have some variation of edge density.  But not too much!  These graphon graphs are always going to have almost all vertices with degree linear in n.  That’s not at all like the networks you encounter in real life, which are typically sparse (vertex degrees growing sublinearly in n, or even being constant on average) and typically highly variable in degree (e.g. degrees following a power law, not living in a band of constant multiplicative width.)  The new theory of L^p graphons is vastly more general.  I’ve only looked at this paper for a half hour but I feel like it’s the answer to a question that’s always bugged me; what are the right descriptors for the kinds of random graphs that actually occur in nature?  Very excited about this, will read it more, and will give a SILO seminar about it on February 4, for those around Madison.
• Wait, I’ve got still one more thing about random graphs!  Russ Lyons gave a plenary about his work with Angel and Kechris about unique ergodicity of the action of the automorphism group of the random graph.  Wait, the random graph? I thought there were lots of random graphs!  Nope — when you try to define the Erdös-Renyi graph on countably many vertices, there’s a certain graph (called “the Rado graph”) to which your random graph is isomorphic with probability 1!  What’s more, this is true — and it’s the same graph — no matter what p is, as long as it’s not 0 or 1!  That’s very weird, but proving it’s true is actually pretty easy.  I leave it an exercise.
• Rick Kenyon gave a beautiful talk about his work with Aaron Abrams about “rectangulations” — decompositions of a rectangle into area-1 subrectangles.  Suppose you have a weighted directed graph, representing a circuit diagram, where the weights on the edges are the conductances of the corresponding wires.  It turns out that if you fix the energy along each edge (say, to 1) and an acyclic orientation of the edges, there’s a unique choice of edge conductances such that there exists a Dirichlet solution (i.e. an energy-minimizing assignment of a voltage to each node) with the given energies.  These are the fibers of a rational map defined over Q, so this actually gives you an object over a (totally real) algebraic number field for each acyclic orientaton.  As Rick pointed out, this smells a little bit like dessins d’enfants!  (Though I don’t see any direct relation.)  Back to rectangulations:  it turns out there’s a gadget called the “Smith Diagram” which takes a solution to the Dirichlet problem on the graph  and turns it into a rectangulation, where each edge corresponds to a rectangle, the area of the rectangle is the energy contributed by the current along that edge, the aspect ratio of the rectangle is the conductance, the bottom and top faces of the rectangle correspond to the source and target nodes, the height of a face is the voltage at that node, and etc.  Very cool!  Even cooler when you see the pictures.  For a 40×40 grid, it looks like this: