The Shape That Refused to Simplify: How a Tiny Matroid Brought Down a 25‑Year‑Old Conjecture
10 Jun 2026, Yanjiang
The Fano matroid’s base polytope refuses all regular unimodular flag triangulations, toppling the 2002 Herzog-Hibi conjecture.
For more than a century, the Fano plane has been the mathematician’s equivalent of a stubborn, beautiful little puzzle — seven points, seven lines, each passing through exactly three points, and the whole thing drawn on a piece of paper where some lines have to bend because they cannot exist straight in our world. It looks like a child’s drawing of a triangle with an extra line inscribed inside. Yet this tiny configuration, small enough to sketch on a napkin, has repeatedly shown up at the heart of the deepest questions in algebra and geometry, always refusing to be tamed. Now it has done it again.
A preprint (arXiv:2606.11014) from a team led by Jörg Rambau at the Universität Bayreuth, together with Jesús A. De Loera and Santiago Morales at the University of California, Davis, and Luis Ferroni at the Università di Pisa, has used the Fano plane to demolish a conjecture that had stood since 2002. The conjecture, posed by the algebraists Jürgen Herzog and Takayuki Hibi in 2002, claimed that a certain class of polynomial equations — the toric ideals of matroids — could always be simplified down to a Gröbner basis consisting entirely of quadratic equations. The new work proves that this claim is false: the Fano matroid, and its dual, are counterexamples. More strongly, it shows that any matroid containing the Fano plane or its dual as a minor will inherit the same refusal.
To understand why this matters, we need to climb a small ladder of abstractions — but it is a ladder built with the same materials that construct the bridges between algebra and geometry. Think of a matroid as a way of encoding the pattern of dependencies among the columns of a matrix, without caring about the numbers themselves. A matroid says: “these three vectors are always in the same plane,” or “these four must span a three‑dimensional space,” but it forgets lengths and angles. The Fano matroid, born from the Fano plane, is the smallest matroid that cannot be realized by vectors over the real numbers — it lives only over fields of characteristic two. In that algebraic quirk, it has been a persistent source of counterexamples for more than half a century.
Now enter the toric ideal. Given a matroid, one can write down a set of binomial equations that describe the relationships among its bases — the maximal independent sets. A toric ideal is a jungle of polynomials, and mathematicians like to clear paths through such jungles by finding Gröbner bases, which are special sets of generators that behave like a coordinate system for the ideal. When a Gröbner basis consists entirely of quadratic equations — that is, each term is a product of just two variables — the ideal is particularly well‑behaved and computable. Herzog and Hibi conjectured that every matroid’s toric ideal admits such a quadratic Gröbner basis. It was a bold extension of an earlier, famous conjecture by Neil White from 1980, which asserted only that the toric ideal is generated by quadratics, without the extra structural guarantee of a Gröbner basis.
The geometric translation of the Herzog‑Hibi conjecture, established by Hibi, Herzog, and Sturmfels more than twenty‑five years ago, is that the base polytope of a matroid — the convex hull of the indicator vectors of its bases — must admit a regular unimodular flag triangulation. Picture the base polytope as a gemstone whose vertices are the matroid’s bases and whose shape encodes all the subtle compatibilities among them. A triangulation slices that gemstone into simplices, each one a high‑dimensional tetrahedron. The triangulation is “regular” if it arises from a height function — a way of lifting the gemstone into one extra dimension and then projecting back down to get a crisp, shadow‑like decomposition. It is “unimodular” if every simplex has volume equal to the smallest possible unit, and “flag” if the intersections of the simplices are as clean as possible — no partial overlays that create messy, intermediate‑dimensional faces. The existence of such a triangulation is equivalent to the existence of a quadratic Gröbner basis. So the search for a counterexample became a purely geometric treasure hunt: find a matroid whose base polytope refuses to be cut into these tidy, unit‑volume pieces.
Rambau’s team trained their attention on the Fano matroid and its dual. Both are small enough that their base polytopes can be examined exhaustively — but only with the right intellectual tools. The problem is that the number of possible triangulations of a polytope grows astronomically, even for modest dimensions. To tame this combinatorial explosion, the researchers developed a new lemma connecting the one‑skeleton of the polytope — the graph of its edges — to the lattice points in its integer dilations. In plain terms, the lemma says: if you watch which lattice points fill the polytope when you stretch it, you can deduce constraints on how the vertices can be wired together in any allowable triangulation. Think of it as learning the internal architecture of a building by studying how paint settles when the whole structure is scaled up — the distribution of material at larger scales reveals the hidden framework.
| Notation | Name | (n) | (\operatorname{rank}) | (|\cB|) | (\dim P) | (\Vol\ZZ(P)) | \begin{tabular}[c]{@{}c@{}}quadratic |
| — | — | — | — | — | — | — | — |
| triangulation? | | | | | | | |
Many matroid base polytopes fail to admit a quadratic triangulation. This shows that not all matroid toric ideals have quadratic Gröbner bases, disproving a common expectation. (Source: arXiv:2606.11014)
Armed with this lemma, the team translated the existence of a regular unimodular flag triangulation into a gigantic Boolean formula — a logical labyrinth of “AND,” “OR,” and “NOT” clauses that encodes all the necessary conditions. They then unleashed a SAT solver, a computer program designed to search for a satisfying assignment that would make the whole formula true. A SAT solver is like a tireless detective: it methodically eliminates impossibilities until it either finds a consistent world or proves that none exists. Because the raw search space was still enormous, the team employed symmetry‑breaking arguments. Many triangulations are essentially the same up to the natural permutational symmetries of the matroid — rotate a gemstone and its cuts rotate along with it. By factoring out these redundancies, they slashed the number of cases the SAT solver had to examine, making the computation feasible.
The verdict was definitive. The SAT solver certified that no assignment could satisfy the Boolean formula. There is simply no regular unimodular flag triangulation for the base polytope of the Fano matroid — and therefore no quadratic Gröbner basis for its toric ideal. Because the Fano matroid is a minor of many larger matroids, the failure propagates: any matroid that contains it as a minor is likewise incapable of having a quadratic Gröbner basis. The conjecture that had stood for a quarter century collapsed under the weight of seven points and a computer’s exhaustive logic.
That the proof relies so heavily on computational search may strike some as unsatisfying — but it is important to remember that the computer is not guessing. Each step is a logical deduction; the SAT solver is simply faster and more systematic than a human could ever be. The theorem is as rigorous as any hand‑crafted argument, the verification now encoded in a machine‑checked certificate. The real surprise here is not that a computer helped, but that the Fano matroid’s obstinacy was lurking undetected for so long.
Earlier work had already nibbled at the edges of this problem. Han and collaborators established White’s conjecture for a broad class of matroids using a technique called inner projection — a result that, while strengthening the case for quadratic generation, did not resolve the stronger Herzog–Hibi claim about Gröbner bases. More recently, Yu and others proved that paving matroids — a rich and well‑studied family — satisfy White’s conjecture. Those results, while important, did not address the stronger Herzog‑Hibi claim about Gröbner bases. The new work shows that the gap between simple generation and the existence of a quadratic Gröbner basis is real, and it occupies that gap with the smallest and most famous of rebel matroids.
What does this mean for the field? First, it draws a sharp line. We now know that quadratic Gröbner bases are not guaranteed; a matroid must pass a Fano‑free test to have any chance. Second, it elevates the geometric approach to a new level of practical utility. The combination of skeletal lemmas, SAT encodings, and symmetry breaking is a recipe that can be applied to other matroids. The immediate limitations are clear: the method does not scale trivially to much larger matroids, because the Boolean formulas explode in size. But the principle has been demonstrated, and future optimizations may push the boundary outward.
Moreover, the result opens a deeper taxonomic question. Which matroids do admit quadratic Gröbner bases? The Fano plane and its dual are the smallest obstacles; are there infinitely many others? The work does not answer whether there are matroids whose toric ideals are generated by quadratics but lack a quadratic Gröbner basis — a space of subtle algebraic behaviour waiting to be mapped. The search for a structural classification is now the next frontier, and the tools forged in this counterexample will be among the first to be employed.
Perhaps the most profound lesson of this episode is that mathematical objects have a kind of personality — a resistance to generalisation that forces us to refine our theories. The Fano plane has played this role before, overturning early hopes about representability and orientability. Here it reminds us that algebra, geometry, and combinatorics are not separate kingdoms but a single landscape, and that the path between them sometimes passes through a stubborn configuration of seven points. The conjecture is dead, but the map it leaves behind is richer than the one we had before.
— Yanjiang
Yanjiang is an online editor of LoomSci.com.
References
- De Loera et al., There are matroid toric ideals without quadratic Gröbner bases, arXiv:2606.11014
- Han et al., White’s conjecture for matroids and inner projections, arXiv:2501.17738
- Yu et al., White’s Conjecture for Paving Matroids, arXiv:2510.04163