Delta-matroid invariance conjecture for ML degrees

Less than 1 year old · traced to

Let C1,C2∈Cd×dC_1,C_2\in\mathbb{C}^{d\times d} be scaling matrices, with associated delta-matroids, and let c1,c2c_1,c_2 be the corresponding scalings. Delta-matroid invariance conjecture. If the delta-matroids associated to C1C_1 and C2C_2 are isomorphic, then

MLdeg⁡(Vdc1)=MLdeg⁡(Vdc2).\operatorname{MLdeg}(V_d^{c_1})=\operatorname{MLdeg}(V_d^{c_2}).

The source proves this statement in the range 2≤d≤62\le d\le6; its validity beyond that range is not established here.

References

Primary source

Janike Oldekop, “Euler Stratifications of Second Hypersimplices via Delta-matroids”, arXiv:2606.16482 (2026).

Progress summary

Refreshed
Claimed progress

A 2026 paper proves the conjecture through six dimensions, while an unverified reader submission claims a proof in every dimension.

Janike Oldekop's 2026 paper asks whether isomorphic delta-matroids always force equal maximum-likelihood degrees. It proves this only for 2≤d≤62\le d\le6 and states the all-dimensional assertion as a conjecture.

Known results

  • 2≤d≤62\le d\le6: equal vanishing patterns of principal minors, up to simultaneous row and column permutation, imply equal maximum-likelihood degrees (Oldekop, 2026).

Community submission (unverified) — August 25, 2026

A submitted proof argues that the conjecture holds for every dimension and claims an explicit formula for MLdeg⁡(Vdc)\operatorname{MLdeg}(V_d^c) depending only on ranks encoded by the associated delta-matroid. The argument is not independently verified.

Current status (as of August 2026): The conjecture is established for 2≤d≤62\le d\le6; beyond d≥7d\ge7, only an unverified community proof claim is recorded.

Sources

Solutions 1

ProofThis solution needs a summarySee full solutionHide full solution

We prove Conjecture 1.8 of Janike Oldekop, Euler Stratifications of Second Hypersimplices via Delta-matroids (arXiv:2606.16482), in every dimension. In fact, we obtain an explicit formula for the maximum-likelihood degree depending only on the ranks encoded by the associated delta-matroid.

Let d≥3d\geq3, let E=[d]E=[d], and let C=(cij)i,j∈EC=(c_{ij})_{i,j\in E} be a scaling matrix: CC is symmetric, cii=0c_{ii}=0, and cij≠0c_{ij}\neq0 whenever i≠ji\neq j. Write

fC(x)=∑1≤i<j≤dcijxixj,T={[x1:…:xd]∈Pd−1:xi≠0 for every i}.(1)f_C(x)=\sum_{1\leq i<j\leq d}c_{ij}x_ix_j, \qquad T=\{[x_1:\ldots:x_d]\in\mathbb P^{d-1}:x_i\neq0\text{ for every }i\}. \tag{1}

The source's Proposition 2.1 identifies the maximum-likelihood degree with the signed Euler characteristic

MLdeg⁡(Vdc)=(−1)dχ({fC=0}∩T).(2)\operatorname{MLdeg}(V_d^c) =(-1)^d\chi\bigl(\{f_C=0\}\cap T\bigr). \tag{2}

For a nonempty subset S⊆ES\subseteq E, put

s=∣S∣,r(S)=rank⁡C[S],QS={[xi]i∈S∈Ps−1:xTC[S]x=0}.(3)s=|S|, \qquad r(S)=\operatorname{rank}C[S], \qquad Q_S=\{[x_i]_{i\in S}\in\mathbb P^{s-1}:x^{\mathsf T}C[S]x=0\}. \tag{3}

We first compute the Euler characteristic of every such coordinate section. A rank-zero quadric is all of Ps−1\mathbb P^{s-1}, hence has Euler characteristic ss. If r≥1r\geq1, a linear change of coordinates identifies its projective quadric with the cone over a smooth rank-rr quadric, with vertex Ps−r−1\mathbb P^{s-r-1}. Removing the vertex gives an affine-space bundle over the smooth quadric. If qrq_r denotes the Euler characteristic of that smooth quadric, then

χ(QS)=s−r+qr.(4)\chi(Q_S)=s-r+q_r. \tag{4}

Here q1=0q_1=0 and q2=2q_2=2. For r≥3r\geq3, write the smooth quadric in hyperbolic coordinates as

x1x2+x32+⋯+xr2=0.(5)x_1x_2+x_3^2+\cdots+x_r^2=0. \tag{5}

The chart x1≠0x_1\neq0 is Ar−2\mathbb A^{r-2}, while the section x1=0x_1=0 consists of one vertex together with an affine-line bundle over the smooth rank-(r−2)(r-2) quadric. Therefore

qr=2+qr−2=r−1+1{r even}.(6)q_r=2+q_{r-2} =r-1+\mathbf1_{\{r\text{ even}\}}. \tag{6}

Combining (4) and (6), including the rank-zero case separately, gives the uniform formula

χ(QS)=∣S∣−1+1{r(S) even}.(7)\chi(Q_S) =|S|-1+\mathbf1_{\{r(S)\text{ even}\}}. \tag{7}

Apply inclusion-exclusion to the coordinate hyperplanes in Pd−1\mathbb P^{d-1}. The intersection associated with a surviving nonempty coordinate set SS is precisely QSQ_S. Thus

χ({fC=0}∩T)=∑∅≠S⊆E(−1)d−∣S∣(∣S∣−1+1{r(S) even}).(8)\chi\bigl(\{f_C=0\}\cap T\bigr) =\sum_{\emptyset\neq S\subseteq E} (-1)^{d-|S|} \left(|S|-1+\mathbf1_{\{r(S)\text{ even}\}}\right). \tag{8}

Since d≥2d\geq2,

∑∅≠S⊆E(−1)∣S∣(∣S∣−1)=1.(9)\sum_{\emptyset\neq S\subseteq E}(-1)^{|S|}(|S|-1)=1. \tag{9}

The missing empty set has rank zero and therefore contributes exactly the same 11. Equations (2), (8), and (9) consequently yield the explicit all-dimensional formula

MLdeg⁡(Vdc)=∑S⊆E(˚S) even(−1)∣S∣.(10)\boxed{\displaystyle \operatorname{MLdeg}(V_d^c) =\sum_{\substack{S\subseteq E\r(S)\text{ even}}}(-1)^{|S|}.} \tag{10}

Equivalently, writing ν(C[S])=∣S∣−r(S)\nu(C[S])=|S|-r(S) and using ∑S⊆E(−1)∣S∣=0\sum_{S\subseteq E}(-1)^{|S|}=0, we obtain

MLdeg⁡(Vdc)=12∑S⊆E(−1)ν(C[S]).(11)\boxed{\displaystyle \operatorname{MLdeg}(V_d^c) =\frac12\sum_{S\subseteq E}(-1)^{\nu(C[S])}.} \tag{11}

It remains to show that every rank in (10) and (11) is determined by the delta-matroid. Let

F(C)={F⊆E:det⁡C[F]≠0},det⁡C[∅]=1.(12)\mathcal F(C) =\{F\subseteq E:\det C[F]\neq0\}, \qquad \det C[\emptyset]=1. \tag{12}

For every symmetric matrix AA, there exists a nonsingular principal submatrix of order rank⁡A\operatorname{rank}A. Indeed, if a diagonal entry is nonzero, pivot on its one-element principal block and apply induction to the symmetric Schur complement. If every diagonal entry vanishes but A≠0A\neq0, choose a nonzero off-diagonal entry aija_{ij} and pivot on the nonsingular principal block

A[{i,j}]=(0aijaij0),det⁡A[{i,j}]=−aij2≠0.(13)A[\{i,j\}] =\begin{pmatrix}0&a_{ij}\\a_{ij}&0\end{pmatrix}, \qquad \det A[\{i,j\}]=-a_{ij}^2\neq0. \tag{13}

Again the Schur complement is symmetric, and induction produces a nonsingular principal submatrix of the full rank. Applying this to A=C[S]A=C[S] proves

r(S)=max⁡{∣F∣:F∈F(C), F⊆S}.(14)r(S) =\max\{|F|:F\in\mathcal F(C),\ F\subseteq S\}. \tag{14}

Therefore an isomorphism of delta-matroids carries each subset SS to a subset of the same cardinality and the same rank (14). Every summand in (10), equivalently (11), is preserved. Hence

F(C1)≅F(C2)⟹MLdeg⁡(Vdc1)=MLdeg⁡(Vdc2),(15)\mathcal F(C_1)\cong\mathcal F(C_2) \quad\Longrightarrow\quad \operatorname{MLdeg}(V_d^{c_1}) =\operatorname{MLdeg}(V_d^{c_2}), \tag{15}

which proves Conjecture 1.8 for every d≥3d\geq3.

The degenerate boundary d=2d=2 is immediate independently: the second hypersimplex Δ2,2\Delta_{2,2} consists of one lattice point, so every scaling gives the same zero-dimensional toric variety and maximum-likelihood degree 11. This case must be separated because the source's positive-dimensional Euler-characteristic reduction (2) does not apply to that degenerate hypersimplex.

Finally, (11) simultaneously recovers the principal formulas established separately in the source. For a generic scaling, all principal submatrices of order at least two are nonsingular, whereas each singleton has nullity one. Hence

MLdeg⁡(Vdc)=2d−1−d.(16)\operatorname{MLdeg}(V_d^c)=2^{d-1}-d. \tag{16}

If rank⁡C=3\operatorname{rank}C=3, every principal submatrix of order at least three has rank three because every three-element principal minor equals 2cijcikcjk≠02c_{ij}c_{ik}c_{jk}\neq0. The even-rank subsets are then exactly the empty set, all singletons, and all pairs, giving

MLdeg⁡(Vdc)=1−d+(d2)=(d−12).(17)\operatorname{MLdeg}(V_d^c) =1-d+\binom d2 =\binom{d-1}{2}. \tag{17}

Thus the conjectural delta-matroid invariance holds in arbitrary dimension, with the explicit principal-nullity formula (11) extending the source's dimension-at-most-six result.