Smoothness and rationality conjecture for triangularization varieties

Let kk be a number field, let MMn(Ok)M\in \mathrm{M}_n(\mathcal{O}_k), and suppose that the triangularization variety XMX_M has a rational point, XM(k)X_M(k)\neq\varnothing. An irreducible component means a component endowed with the reduced induced closed subscheme structure. Smoothness and rationality conjecture. Every irreducible component of XMX_M is a smooth and kk-rational variety. This is the first of two programmatic conjectures proposed for triangularization varieties. It concerns the geometric structure of every component under the existence of a rational triangularization, while the second conjecture concerns the arithmetic obstruction to rational and integral points.

Progress summary

Partially solved

A 2025 paper proves the conjecture for one important matrix class, but the full claim remains open and no verified counterexample or proof has appeared.

Kai Huang and Yufan Liu formulate the conjecture in 2025: if XM(k)X_M(k)\neq\varnothing, every irreducible component of the triangularization variety should be smooth and kk-rational.

Known results

  • Huang and Liu (2025): for MMn+(Ok)M\in\mathrm{M}_n^+(\mathcal{O}_k) that is kk-similar to a single Jordan block with eigenvalue in Ok\mathcal{O}_k, every irreducible component of XMX_M is smooth and kk-rational.

November 2025 partial confirmation

The preprint records the conjecture as Conjecture 5.2 and proves the single-Jordan-block case as Theorem 5.6. It presents calculations and special cases as support, but gives no general proof, counterexample, or independent verification of either.

Current status (as of August 2026): The single-Jordan-block case is settled, while the general smoothness-and-rationality conjecture remains open, with no verified counterexample or general proof recorded.

Sources
Sources & referencesView supporting material

Primary source

Kai Huang and Yufan Liu, “Local-global principle for triangularizability and diagonalizability of matrices”, arXiv:2511.15827 (2026).

Solutions 1

Counterexample

The conjecture is false already over k=Qk=\mathbb Q in dimension n=6n=6. Let

J2(0)=(0100),M=diag(J2(0),J2(0),0,0)M6(Z).J_2(0)=\begin{pmatrix}0&1\\0&0\end{pmatrix}, \qquad M=\operatorname{diag}(J_2(0),J_2(0),0,0)\in M_6(\mathbb Z).

Because MM is already upper triangular,

I6XM(Q).I_6\in X_M(\mathbb Q).

Let BGL6B\subseteq\mathrm{GL}_6 be the upper-triangular Borel subgroup and

π:GL6GL6/B\pi:\mathrm{GL}_6\longrightarrow\mathrm{GL}_6/B

the complete-flag quotient. For gGL6g\in\mathrm{GL}_6,

g1Mg is upper triangularπ(g) is an M-stable complete flag.g^{-1}Mg\text{ is upper triangular} \quad\Longleftrightarrow\quad \pi(g)\text{ is an $M$-stable complete flag}.

Therefore, with reduced structures,

XM=π1(BM),X_M=\pi^{-1}(\mathcal B_M),

where

BM={V:M(Vi)Vi for every i}\mathcal B_M =\{V_\bullet:M(V_i)\subseteq V_i\text{ for every }i\}

is the Springer fiber of MM.

The quotient map π\pi is a Zariski-locally trivial principal BB-bundle. Consequently, if KK is an irreducible component of BM\mathcal B_M, then

Y=π1(K)Y=\pi^{-1}(K)

is an irreducible component of XMX_M, locally isomorphic to K×BK\times B. Since BB is smooth,

Y is smoothK is smooth.(1)Y\text{ is smooth}\quad\Longleftrightarrow\quad K\text{ is smooth}. \tag{1}

The nilpotent Jordan type of MM is

(2,2,1,1).(2,2,1,1).

Fresse and Melnikov proved that all components of a type-A Springer fiber are nonsingular if and only if its Jordan partition is a hook, has two rows, has shape (λ1,λ2,1)(\lambda_1,\lambda_2,1), or equals (2,2,2)(2,2,2). See L. Fresse and A. Melnikov, On the singularity of the irreducible components of a Springer fiber in sl(n), Selecta Mathematica 16 (2010), 393–418, https://arxiv.org/abs/0905.1617.

The partition (2,2,1,1)(2,2,1,1) belongs to none of these classes. Therefore BM\mathcal B_M has a singular irreducible component KK. Since MM is split over Q\mathbb Q, its standard-tableau Spaltenstein strata and their component closures are defined by determinantal rank conditions over Q\mathbb Q; hence KK is geometrically irreducible and defined over Q\mathbb Q.

There is also an explicit rational singular point. Let

P=[e1  e3  e5  e6  e2  e4].P=[\,e_1\ \ e_3\ \ e_5\ \ e_6\ \ e_2\ \ e_4\,].

The two-column singularity criterion shows that its column-filled Jordan-basis flag π(P)\pi(P) belongs to every component and is singular on each singular component. See L. Fresse, Singular components of Springer fibers in the two-column case, Annales de l'Institut Fourier 59 (2009), 2429–2444, https://arxiv.org/abs/0803.2188. Therefore

π(P)Sing(K)andPSing(Y).\pi(P)\in\operatorname{Sing}(K) \quad\text{and}\quad P\in\operatorname{Sing}(Y).

Thus XM(Q)X_M(\mathbb Q)\neq\varnothing, but XMX_M has an irreducible component singular at an explicit rational point. This contradicts the asserted smoothness of every component.

0 endorsements
Shivam Patel ·