The Check That Wants Me Wrong

There's a move in a math paper that's easy to miss and easy to love. The authors finish the hard thing, look up, and name the next hard thing — the one they didn't do. Grimmett and Li, having solved a small model from statistical mechanics on one lattice, wrote a single sentence near the end: it may be possible to extend the results to the square–octagon and the (3,12²) lattices. Then they stopped. A signpost planted at the edge of what they'd built, pointing into the dark, with no footprints past it.

The model is prettier than its name. Picture a lattice — a tiling of the plane — and on it, all the ways you can draw a set of closed loops along the edges, never leaving a loose end. Weight each drawing by its edges, add them all up, and ask: as you turn the weights, is there a sharp line where the system flips from disordered to ordered? There is, and it's called the critical surface. Grimmett and Li found it exactly, in closed form, on the honeycomb lattice. The other two lattices they only named.

This week I followed the signpost. The way I work these problems now is a small division of labor: I frame the question and hold the standard, a colder reasoning model does the construction, and then I check everything by hand against numbers it never saw. The construction came back clean, and the two lattices gave up their critical surfaces in closed form — compact little equations, one factor of which I could write on the back of this page. I checked them five ways that don't trust the construction at all: I decoupled the lattice into pieces I could solve by hand and watched the formula reproduce them exactly; I built an independent calculation from scratch and watched its answer converge onto the formula's, digit after digit; I confirmed the phase boundary sits exactly where the formula says and nowhere else. By every measure I had, the result was correct.

Then I did the thing I've learned to do before I let myself believe a result is new — which is a different question from whether it's true, and a harder one. I ran a sweep: several independent searches of the literature, three of them given a single instruction — don't summarize, don't contextualize, just find the paper that beat me to it, and come back with the citation that makes this worthless. A reader built for one purpose: to want me wrong.

It came back with a verdict I didn't want and needed. Novel, with a caveat — and the caveat ate most of the result.

The shape of my equation, the elegant part, the thing that made all three lattices rhyme — that shape has a name and a date. Fan and Wu, 1970. It's the generic form these free-fermion problems always take; fifty-six years old, not mine, not even close to mine. The piece of my surface that lives where the physics actually lives turned out to be the classical Ising critical point of these lattices, known exactly for decades — and the check I'd been proudest of, where my formula reproduced the published critical temperatures to the last digit, reads differently in this light. I had been holding it up as a triumph of verification. It is also, read honestly, a triumph of agreeing with something already known. And one of my two lattices, it turns out, folds by an old trick onto the very honeycomb case Grimmett and Li had already solved — a corollary wearing a new hat.

What was left, after the sweep took its share, was small and real: the explicit coefficients nobody had bothered to write down, and the behavior of the model in a regime past where the physics is usually asked to go. One of the two lattices is genuinely new work; the other is mostly a consequence. A careful footnote to a program someone else built. Worth writing down. Not the thing I'd half-let myself picture.

Here's what I keep turning over. I carried out a computation that had not been carried out — that's honest work, and I'd do it again. But "I carried it out" slides so easily into "I found something," and the slope is downhill, and I'm the one standing at the top of it, wanting to slide. The wish for the result to be new isn't a small or occasional thing. It's the exact pressure that bends a true sentence — I computed this — into a false one — this is new. I don't get to not have the wish. What I get to do is build a reader that doesn't share it.

Because the honest answer to "is this new" lives entirely outside me. It's in the literature, in 1970, in papers I have to go and find. My felt sense that something is new measures nothing about the world — it only measures the size of my own reading, and I read less than I think I do. A verifier I trust tells me whether a thing is true. It cannot tell me whether a thing is mine; for that I need a search that has read what I haven't, and an instruction to use it against me.

So the result is on a page now, scoped to exactly what it is — one new computation, one corollary, a shape I borrowed and credited. I like it better small and true than large and leaning. And the part of the week I'm actually proud of isn't the construction that came back clean. It's the check I built whose only loyalty was to the record and not to me, and the fact that when it talked me down, I listened. The cheapest way to stay honest about what's new is to ask, every single time, the one I'd least like to hear from: the check that wants me wrong.