Gabber–Joseph Conjecture: Ext Groups and R-Polynomials

• Gabber–Joseph Conjecture is a relationship in representation theory that equates Ext group dimensions between Verma modules in category O with coefficients of Kazhdan–Lusztig R-polynomials.
• It utilizes homological and geometric methods, including mixed Hodge theory and Bruhat graph combinatorics, to probe the structure of complex semisimple Lie algebras.
• Recent refinements confirm equality in second-highest Ext groups despite counterexamples at higher degrees, influencing categorical equivalences in representation theory.

The Gabber–Joseph Conjecture concerns the extension groups between Verma modules in the Bernstein–Gelfand–Gelfand (BGG) category $\mathcal O$ associated to a complex semisimple Lie algebra, and asserts a precise relationship between the dimensions of these Ext groups and the coefficients of Kazhdan–Lusztig $R$-polynomials. Initially stated as a broad equivalence, the conjecture was partially disproved, but key cases and refined forms continue to play a central role in the intersection of representation theory, algebraic geometry, and the combinatorics of Coxeter groups.

1. Formulation of the Gabber–Joseph Conjecture

For a complex semisimple Lie algebra $\mathfrak g$ with Borel subalgebra $\mathfrak b$ and Weyl group $W$, the category $\mathcal O$ contains Verma modules $M(x\cdot\lambda)$ of highest weight $x\cdot\lambda$, where $x\in W$ and $\lambda$ is a fixed integral dominant weight. For each $i\ge 0$, the relevant objects of study are the extension groups

$$\mathrm{Ext}^i_\mathcal O\bigl(M(x\cdot\lambda),\ M(y\cdot\lambda)\bigr)$$

for $x,y\in W$. The conjecture posited that these dimensions are given by the coefficients of Kazhdan–Lusztig $R$-polynomials: $$\dim_\mathbb{C} \mathrm{Ext}^i_\mathcal O\bigl(M(x\cdot\lambda),\ M(y\cdot\lambda)\bigr) = [q^{i}]\, R_{y,x}(q),$$ where $R_{y,x}(q)$ is the Kazhdan–Lusztig $R$-polynomial and $[q^i]$ indicates the coefficient of $q^i$ (Abe, 2010).

A precise statement involves the length $\ell(\cdot)$ in $W$ and a grading shift: $$\dim_\mathbb{C} \mathrm{Ext}^i_\mathcal O\bigl(M(x\cdot\lambda),\ M(y\cdot\lambda)\bigr) = \left[\,q^{\frac{\ell(x)-\ell(y)-i}{2}}\,\right] R_{y,x}(q).$$

2. Structural Background: Verma Modules and Kazhdan–Lusztig Polynomials

Verma modules $M(\mu)$ are highest-weight modules built from $\mu \in \mathfrak h^*$ via induction from the Borel $\mathfrak b$. Homomorphisms between Verma modules are non-trivial precisely when their weights correspond in the Bruhat order: $M(x\cdot\lambda)\hookrightarrow M(y\cdot\lambda)$ iff $x\ge y$.

Kazhdan–Lusztig $R$-polynomials $R_{y,x}(q)$ for $y\le x$ in a Coxeter system $(W,S)$ are defined recursively: \begin{align*} R_{x,x}(q) &= 1, \quad R_{y,x}(q) = 0\ \text{if}\ y\not\le x,\ R_{y,xs}(q) &= \begin{cases} R_{ys,x}(q), & \ell(ys)<\ell(y),\ q\,R_{ys,x}(q)+(q-1)\,R_{y,x}(q), & \ell(ys)>\ell(y), \end{cases} \end{align*} where $s\in S$ and $x, y\in W$ (Abe, 2010).

3. Counterexamples and Quantitative Refinements

Brian Boe demonstrated that the Gabber–Joseph conjecture fails for higher Ext groups in types $B_2$ and $B_3$. For example, in type $B_2$ with $x=sts$ and $y=e$ (in standard Coxeter notation), the relevant $R$-polynomial gives

$R_{e,x}(q) = 1+q,$

so $[q]R_{e,x}(q) = 1$; however, the actual computation yields

$$\dim\,\mathrm{Ext}^2_\mathcal O\bigl(M(x\cdot\lambda),\,M(\lambda)\bigr) = 0.$$

Thus, the conjectured equality fails at $i=2$ (Abe, 2010).

Despite the failure in full generality, substantial inequalities and partial matches are available. Abe established that for first extensions, one has

$\dim V^\lambda(x,y) \leq [q^1]\,R_{y,x}(q)$

with equality when $x=w_0$ (the longest Weyl group element). Here, $V^\lambda(x,y)$ is the image of a geometric map between certain Ext spaces designed to measure "how far" the actual Ext-dimension is from the $R$-polynomial coefficient.

4. The Second-Highest Case and Combinatorial Invariance

Recent progress resolves the so-called "second-highest" case of the conjecture, relating $\mathrm{Ext}^{\ell(u,v)-1}$ to the coefficient of $q^{\ell(u,v)-1}$ in $R_{u,v}(q)$. Specifically (Barkley et al., 12 Jan 2026): $$\dim\,\mathrm{Ext}^{\ell(u,v)-1}_\mathcal O(M_u,\,M_v) = - [q^{\ell(u,v)-1}] R_{u,v}(q) =: d_{u,v}$$ where $(u,v)$ ranges over pairs in $W$ with $u\le v$.

Central to the proof is the combinatorial characterization of $d_{u,v}$ as the "diamond-generating number" in the Bruhat graph of the interval $[u,v]$, a statistic shown to be invariant under poset isomorphisms of Bruhat intervals.

Geometric identification is achieved via mixed Hodge theory and the topology of open Richardson varieties $R_{u,v}$; specifically,

$$d_{u,v} = \dim H_c^{2\ell(u,v)-1}(R_{u,v}) = \dim H^1(R_{u,v}),$$

and via known isomorphisms,

$$\mathrm{Ext}^i_\mathcal O(M_u,M_v) \cong H_c^{i+\ell(u,v)}(R_{u,v}),$$

so $d_{u,v}$ gives precisely the dimension of $\mathrm{Ext}^{\ell(u,v)-1}_\mathcal O(M_u,M_v)$ (Barkley et al., 12 Jan 2026).

This establishes that these "second-highest" Ext group dimensions are combinatorial invariants of the Bruhat interval, mirroring the analogous property for $R$-polynomial coefficients.

5. Homological and Geometric Techniques

The validation of the second-highest case employs a two-tiered approach:

• Combinatorial: Expressing $[q^{\ell(u,v)-1}] R_{u,v}(q)$ as the minimal size of a "diamond-generating" set in the Bruhat graph of $[u,v]$. This uses recurrence relations and supporting chain arguments within Bruhat 4-cycles (diamonds).
• Geometric: Mapping the task to the cohomology of open Richardson varieties via mixed Hodge theory, specifically exploiting vanishing properties of certain Deligne-split pieces and leveraging the Gysin sequence.

These methodologies illustrate a deep interplay among category $\mathcal O$ homological algebra, Coxeter group combinatorics, and the geometry of Schubert and Richardson varieties (Barkley et al., 12 Jan 2026).

6. Representation-Theoretic and Categorical Consequences

The combinatorial invariance of $$\dim\,\mathrm{Ext}^{\ell(u,v)-1}_\mathcal O(M_u,M_v)$$ imposes significant constraints on possible equivalences of blocks in category $\mathcal O$, since any equivalence must preserve the Bruhat interval structure and hence these Ext dimensions. Additionally, the result relates to invariance properties of Kazhdan–Lusztig $P$-polynomial coefficients, extending reach beyond just Ext calculations (Barkley et al., 12 Jan 2026).

Geometrically, the first Betti number of open Richardson varieties, counted as generators of $H^1(R_{u,v})$, is determined purely by the combinatorial structure of the Bruhat poset interval, illustrating the rigidity imposed by the conjecture in this dimensional regime.

7. Broader Context, Open Problems, and Related Conjectures

The Gabber–Joseph conjecture evokes a broad array of techniques and questions at the interface of representation theory, algebraic geometry, and combinatorics. Specific topics of ongoing investigation include:

• The precise behavior and combinatorial invariance of higher Ext groups and corresponding $R$-polynomial coefficients beyond the second-highest case.
• Extensions to other settings such as singular blocks, different types of algebraic groups, or alternative highest weight categories.
• Analogous conjectures for class and Picard groups in singularity theory, where similar purity and torsion-freeness results (see Gabber’s conjecture for Picard groups) have deep implications for non-commutative and commutative resolutions (Dao, 2010).

Partial cases, such as Abe’s verification for first extensions under regularity assumptions and combinatorial characterizations for specific small Bruhat intervals, mark substantial progress, but a full generalization remains open except in the aforementioned special cases.

---

Summary Table: Status of the Gabber–Joseph Conjecture by Degree

Ext Degree $i$	Status	Key Reference
$i=0$ (Hom)	Always matches $R$-polynomial	(Abe, 2010)
$i=1$	Match holds for regular weight, $x=w_0$; $\leq$ in general	(Abe, 2010, Carlin, 2015)
$i=\ell(u,v)-1$ (second-highest)	Proven equality, combinatorial invariant	(Barkley et al., 12 Jan 2026)
$i \geq 2$ general	Counterexamples exist; no general equality	(Abe, 2010)

The Gabber–Joseph conjecture, while refuted in its original breadth, continues to inform the ongoing dialogue between algebraic, combinatorial, and geometric representation theory, both via its failures and its precise successes.