Capítulo Ju · Vol VII · Scientist Gallery
C K F = build the right structure U

June Huh

He proved that certain sequences of numbers must rise and then fall, by inventing a geometry for objects that have none. His field is the nearest neighbour this corpus has — near enough that reading him cost us three sentences.

b. 1983 · Fields Medal 2022 · Princeton · Hodge theory where there is no space

1 · The late arrival

Huh wanted to be a poet. He dropped out of high school, wrote, and came to mathematics late — studying physics and astronomy at Seoul National University before a visiting professor changed the direction of his life. That professor was Heisuke Hironaka, himself a Fields Medallist, who won the medal for resolution of singularities in characteristic zero.

The detail matters for this gallery, and not for the sentimental reason. Hironaka's subject is singularities — the points where a smooth object folds, pinches, or fails to be a manifold. Huh learned his trade from the person who taught mathematics how to take a singular object apart and rebuild it as a smooth one. What he then did with that training was to carry it somewhere no one expected it to go: into combinatorics, where there are no spaces at all.

He took his PhD at Michigan in 2014, and the Fields Medal in 2022, the first of Korean descent. The citation reads: “for bringing the ideas of Hodge theory to combinatorics, the proof of the Dowling–Wilson conjecture for geometric lattices, the proof of the Heron–Rota–Welsh conjecture for matroids, the development of the theory of Lorentzian polynomials, and the proof of the strong Mason conjecture.”

2 · What he actually proved

A sequence of positive numbers $a_0, a_1, \ldots, a_r$ is log-concave when $a_k^2 \ge a_{k-1}a_{k+1}$ for every interior $k$. Log-concave sequences rise, peak once, and fall. They do not wobble.

Read's conjecture · 2012
The coefficients of a chromatic polynomial are log-concave

Colour the vertices of a graph with $q$ colours so no edge joins two of the same. The number of ways is a polynomial in $q$ — the chromatic polynomial. Its coefficients alternate in sign, and their absolute values had been conjectured since 1968 to be log-concave. Huh proved it.

The route was the surprise: he read the coefficients as characteristic classes on an algebraic variety built from the graph, and applied inequalities that hold for varieties.

Adiprasito – Huh – Katz · Annals 2018
Heron–Rota–Welsh: the same, for all matroids

A matroid is the abstraction of independence — the common skeleton of linearly independent vectors, acyclic edge sets, and algebraically independent elements. Most matroids are not representable: no vectors realise them, so there is no variety to borrow inequalities from. The 2012 method had nowhere to stand.

So they built the standing place. For an arbitrary matroid they constructed a ring that behaves like the cohomology of a smooth projective variety — Poincaré duality, the hard Lefschetz theorem, and the Hodge–Riemann relations — and proved those properties combinatorially, with no space underneath. Log-concavity then falls out of Hodge–Riemann in degree one.

Brändén – Huh · Annals 2020, 192(3)
Lorentzian polynomials

A class of homogeneous polynomials sitting between stable polynomials and volume polynomials of convex bodies, characterised by elementary linear algebra — the Hessian has exactly one positive eigenvalue. Matroids, and more generally M-convex sets, are characterised by the Lorentzian property. The strong Mason conjecture from 1972 follows.


3 · Why this is close to home

Four adjacencies, in increasing order of discomfort.

Their objectOursDistance
Characteristic polynomial of a matroid Characteristic polynomial of the n-bonacci recurrence, $x^n - x^{n-1} - \cdots - 1$ Same words. Different objects — see §4.
Singularity theory via Hironaka The Whitney $A_1$ fold at $q=1$, $V''(1) = 6 \neq 0$ Same vocabulary, and it is their vocabulary.
Tropical geometry, Bergman fans Chapter Tr — max-plus semiring, the tropicalised ladder converging to $\tau = 2$ Genuinely the same machinery.
Cohomology ring with Poincaré duality, built by hand $\chi(H^*(X^6)) = 33$ — Vol VI's central open conjecture An Euler characteristic of a cohomology ring. This is his street.

The last row is the one to sit with. Vol VI's open conjecture is a statement about the cohomology of an object, and the person who most recently taught mathematics how to prove such statements for objects that are not varieties is alive, at Princeton, and has a Fields Medal for exactly that. If $X^6$ has a Chow-ring presentation, the tools to attack $\chi = 33$ already exist and are not ours.

4 · The boundary, computed

Adjacency is not overlap, and the difference should be established rather than asserted. Two computations settle it.

Proposition · verified symbolically, SymPy
The n-bonacci polynomial is not a matroid characteristic polynomial

For any matroid $M$ of positive rank on a loop-free ground set, $\chi_M(1) = 0$: the characteristic polynomial always vanishes at $q=1$. This is standard and is the reason the reduced characteristic polynomial $\chi_M(q)/(q-1)$ is the object Huh's theorems actually govern.

p_n(x) = x^n - x^(n-1) - ... - x - 1 p_2(1) = -1 p_3(1) = -2 p_4(1) = -3 p_5(1) = -4 p_6(1) = -5 -> p_n(1) = 1 - n

Nonzero for every $n > 1$. No n-bonacci polynomial is the characteristic polynomial of any matroid.

Proposition · verified symbolically
Log-concavity holds here, and says nothing
n=2: |coeffs| = (1,1,1) log-concave? yes strictly? no n=3: |coeffs| = (1,1,1,1) log-concave? yes strictly? no n=4: |coeffs| = (1,1,1,1,1) log-concave? yes strictly? no n=5: |coeffs| = (1,1,1,1,1,1) log-concave? yes strictly? no n=6: |coeffs| = (1,1,1,1,1,1,1) log-concave? yes strictly? no the inequality a_k^2 >= a_{k-1} a_{k+1} reads 1 >= 1 at every index

The absolute coefficients of $p_n$ are the constant sequence. Log-concavity is saturated everywhere — true, and empty. Huh's theorems are about Whitney numbers, which grow and then shrink and where the inequality does real work. Ours is the degenerate case.

So the boundary is sharp, and it runs in our favour in one narrow sense: nothing in the n-bonacci ladder is subsumed by, or in competition with, the log-concavity programme. The n-bonacci constants are dominant roots of Pisot type — number theory, not matroid theory. And the dm³ system is a smooth contact flow: Darboux, Reeb fields, limit cycles. There is no matroid anywhere in it. The word appears in this repository only inside the vendored Mathlib.

5 · What reading him cost

It cost three sentences, and they were ours.

The exercise of asking “how would a reader of this literature see the ladder chapters?” is what surfaced the defect audited in WP-61. Several chapters said the dm³ potential $V(q) = q^3 - 3q$ “has a double root at $q=1$”. It does not; $V_3(1) = -2$, and its roots are $0, \pm\sqrt3$. One said the potential $V_\varphi$ has a “degenerate double root” at a point which is neither a root nor degenerate. One claimed $\eta$ is the only n-bonacci constant producing that root, when no n-bonacci constant equals 3 at all.

The underlying mathematics was right in every case. What was wrong was that the prose reached for degenerate when it had a non-degenerate critical point — claiming a deeper singularity than it possessed, while possessing a perfectly good one. In most rooms that would pass. In this one it would not, because these are the words Hironaka's students use to mean specific things.

The lesson, stated as a method When a claim borrows a technical term from a neighbouring field, check the term against that field's meaning, not against your own surrounding argument. An argument can be internally consistent and still be using a word that means something else to everyone who owns it.

6 · What one would actually ask him

In the register of the conference pointer pages: questions, not claims.

Ask · on $X^6$ Vol VI conjectures $\chi(H^*(X^6)) = 33$ for a six-dimensional Wigner-crystal candidate. Is there a presentation of $X^6$ as a Chow ring of a combinatorial object — a fan, a lattice, a matroid? If yes, the AHK machinery applies and the conjecture is a computation rather than a mystery. If no, what obstruction rules it out? Either answer closes an open item.
Ask · on the ladder The n-bonacci polynomials are provably not matroid characteristic polynomials (§4). Is there nonetheless a matroid, or an M-convex set, whose rank sequence indexes the ladder $\pi \to \varphi \to \mu \to \eta \to \Delta \to \Sigma \to \Omega$? The series treats rank $n$ as the structural parameter; matroid theory is the mathematics of rank. If the answer is no, the ladder is analytic and should stop borrowing combinatorial language.
Ask · on method The AHK proof works by constructing the object whose properties you need, rather than manipulating the polynomial you were handed. That is, structurally, the $F$ operator. Is there a general account of when this move is available — when a combinatorial gadget admits a Hodge-theoretic completion — or is each case bespoke?
Open · carried forward

[OPEN] Whether $X^6$ admits a Chow-ring presentation. Unknown; AXLE Issue 6.

[OPEN] The rank $n = 3 \leftrightarrow$ coefficient $c = 3$ correspondence. Arithmetically consistent after the WP-61 repair, but no derivation shows it is more than a coincidence of small integers.

[OPEN] Four ladder files flagged by the WP-61 signature scan and not yet read: chEps-gronwall, chRho-spectral, chH-collatz, chE-gtct.

7 · Where this sits in the AULA track

This chapter is D1–D2 material and does not belong in a first pass. But the ladder it is about is taught from AULA 102 onward, and the correction in §5 lands directly on lesson material, so the route in and out is worth stating.

WhereWhat it coversRelation to this chapter
dm³ 102 · w06 Tribonacci η weighting — the DNLS lab, run as an experiment Corrected 2026-08-12. The lesson placed φ, η and Δ on the same axis as c* = 3, which reads as η = c*. The ladder is indexed by rank n; the threshold is the coefficient c. Both are 3; the constants are not.
Ch η Tribonacci as critical constant — the empirical result behind Milestone III Carried the double-root wording; repaired under WP-61.
dm³ 103 Σ Pentanacci, Ω Hexabonacci — the top of the ladder, D1–D2 The right place to read this chapter, once the ladder is in hand.
Sessão S2 Teorema 2.1 and the asymmetric basin — the Vol IV mini-curso Same discipline applied to a different word: the Grönwall ball is not the basin, and calling it one misclassifies orbits.
Hour House · AULA The lesson programme itself — AULA 101 / 102 / 103 against CEFR Where a student meets the ladder before meeting this chapter.
Why a Fields Medallist belongs in an ESL curriculum Hour House teaches adult English through live research rather than worksheets, and the argument of this chapter is unusually well shaped for that: a claim was made, a neighbouring field's vocabulary was borrowed, the borrowing was checked, and three sentences did not survive. Students at B2–C1 are being asked to justify rather than describe — which is exactly the operation performed here, at professional stakes. The correction is the lesson. It is also, in the operator language the course uses, an F: the structure was rebuilt, not the conclusion discarded.

8 · Coda

The gallery collects people whose work turned out to be the operator chain in another language. Huh is the uncomfortable entry, because his work is not a distant rhyme — it is the adjacent field, doing to combinatorics what this series claims to do to physics, biology and markets: finding that objects with no apparent geometry have one, and that the geometry does the proving.

The difference is that he closed his. The Heron–Rota–Welsh conjecture stood for fifty years and is now a theorem. That is the standard the neighbouring field sets, and it is the right standard to be measured against. The honest position for this corpus is the one it already takes in its own working papers: state the claim precisely enough that it can fail, then check it rather than restate it — and when a neighbour's vocabulary is borrowed, borrow the neighbour's rigour with it.

9 · References

Adiprasito, K., Huh, J., & Katz, E. (2018). Hodge theory for combinatorial geometries. Annals of Mathematics, 188(2), 381–452.
Brändén, P., & Huh, J. (2020). Lorentzian polynomials. Annals of Mathematics, 192(3), 821–891.
Huh, J. (2012). Milnor numbers of projective hypersurfaces and the chromatic polynomial of graphs. Journal of the American Mathematical Society, 25(3), 907–927.
Huh, J., & Katz, E. (2012). Log-concavity of characteristic polynomials and the Bergman fan of matroids. Mathematische Annalen, 354(3), 1103–1116.
International Mathematical Union (2022). Fields Medal citation: June Huh. ICM 2022.

A note on these references Bibliographic details above are given so a reader can find the papers; they have not been re-verified page-by-page against the printed volumes. Where an argument in this chapter depends on what one of these texts actually says — the $\chi_M(1) = 0$ fact in §4, principally — the statement is standard and is also checkable directly from the definition of the characteristic polynomial, which is the route this chapter takes. Nothing here rests on a citation doing work a computation should do.

10 · How to cite this chapter

Citation

Grossi, Pablo Nogueira (2026). June Huh — The Neighbour Who Proves Things. Principia Orthogona, Vol VII · Scientist Gallery, Chapter Ju. G6 LLC, Newark, New Jersey.
totogt.github.io/geometry/book7/ch-huh.html

@incollection{grossi2026huh, author = {Grossi, Pablo Nogueira}, title = {June Huh --- The Neighbour Who Proves Things}, booktitle = {Principia Orthogona, Vol VII: The Scientist Gallery}, chapter = {Ju}, publisher = {G6 LLC}, address = {Newark, New Jersey}, year = {2026}, month = {8}, url = {https://totogt.github.io/geometry/book7/ch-huh.html}, note = {No DOI assigned as of 2026-08-12.} }

On identifiers. No DOI has been assigned to this chapter as of 2026-08-12, and Vol VII has no standalone deposit. There is no series-level DOI. For Principia Orthogona as a whole, cite the Zenodo community zenodo.org/communities/principia-orthogona — never a concept DOI, which resolves to the most recent deposit and therefore pins nothing. Vol VII has no ISBN of its own; per the registry, 979-8-9954416-5-6 is an unallocated reserve number and is not a valid fallback. The works by Huh and coauthors above should of course be cited directly, not through this chapter.

← Geometria Tropical · the shared machinery WP-61 · the repair this prompted →