Table of Contents
Fetching ...

Log-concavity of characters of parabolic Verma modules, and of restricted Kostant partition functions

Apoorva Khare, Jacob P. Matherne, Avery St. Dizier

TL;DR

The paper proves that shifted, normalized characters of parabolic Verma modules over $\mathfrak{sl}_{n+1}$ (and their semisimple products) are Lorentzian, yielding both continuous and discrete log-concavity in type $A$ root directions. It connects parabolic Vermas to restricted Kostant partition functions via flow polytopes and uses Alexandrov–Fenchel inequalities to deduce discrete log-concavity, with a parallel flow-polytope proof. The results show that parabolic Vermas form a maximal class with log-concave characters, while higher-order Verma modules and non-$A$ types fail to preserve these properties, and they extend HMMS by addressing integral and nonintegral weights and direct-sum algebras. The work also provides counterexamples in Jack/Macdonald settings and outside type $A$, clarifying the limits of log-concavity in representation-theoretic and symmetric-function contexts. Overall, the Lorentzian/flow-polytope framework unifies discrete and continuous log-concavity phenomena for a broad family of highest-weight modules and their associated partition functions, with significant implications for Kostant partition functions and Schur-type polynomials.

Abstract

In 2022, Huh-Matherne-Mészáros-St. Dizier showed that normalized Schur polynomials are Lorentzian, thereby yielding their continuous (resp. discrete) log-concavity on the positive orthant (resp. on their support, in type $A$ root directions). A reinterpretation of this result is that the characters of finite-dimensional simple representations of $\mathfrak{sl}_{n+1}(\mathbb{C})$ are denormalized Lorentzian (DL). In the same paper, these authors also showed that shifted characters of Verma modules over $\mathfrak{sl}_{n+1}(\mathbb{C})$ are DL. In this work we extend these results to a larger family of modules that subsumes both of the above: we show that shifted characters of all parabolic Verma modules over $\mathfrak{sl}_{n+1}(\mathbb{C})$ are denormalized Lorentzian. The proof involves certain graphs on $[n+1]$; more strongly, we explain why the character (i.e., generating function) of the Kostant partition function of any loopless multigraph on $[n+1]$ is Lorentzian after shifting and normalizing. We then show that parabolic Vermas form a "maximal" class with log-concave (hence DL) characters. Namely, log-concavity fails in greater generality along three natural directions: (1) it does not hold for every simple Lie type, (2) nor for a larger universal family of highest weight modules, the higher order Verma modules, even in type $A$, and (3) it does not always hold for important generalizations of Schur polynomials: the Jack and Macdonald polynomials. Finally, we extend these results to parabolic (i.e. "first order") and higher order Verma modules over the semisimple Lie algebras $\oplus_{t=1}^T \mathfrak{sl}_{n_t+1}(\mathbb{C})$. We also partially resolve a conjecture of Huh et al on the DL property for integral highest weight simple modules.

Log-concavity of characters of parabolic Verma modules, and of restricted Kostant partition functions

TL;DR

The paper proves that shifted, normalized characters of parabolic Verma modules over (and their semisimple products) are Lorentzian, yielding both continuous and discrete log-concavity in type root directions. It connects parabolic Vermas to restricted Kostant partition functions via flow polytopes and uses Alexandrov–Fenchel inequalities to deduce discrete log-concavity, with a parallel flow-polytope proof. The results show that parabolic Vermas form a maximal class with log-concave characters, while higher-order Verma modules and non- types fail to preserve these properties, and they extend HMMS by addressing integral and nonintegral weights and direct-sum algebras. The work also provides counterexamples in Jack/Macdonald settings and outside type , clarifying the limits of log-concavity in representation-theoretic and symmetric-function contexts. Overall, the Lorentzian/flow-polytope framework unifies discrete and continuous log-concavity phenomena for a broad family of highest-weight modules and their associated partition functions, with significant implications for Kostant partition functions and Schur-type polynomials.

Abstract

In 2022, Huh-Matherne-Mészáros-St. Dizier showed that normalized Schur polynomials are Lorentzian, thereby yielding their continuous (resp. discrete) log-concavity on the positive orthant (resp. on their support, in type root directions). A reinterpretation of this result is that the characters of finite-dimensional simple representations of are denormalized Lorentzian (DL). In the same paper, these authors also showed that shifted characters of Verma modules over are DL. In this work we extend these results to a larger family of modules that subsumes both of the above: we show that shifted characters of all parabolic Verma modules over are denormalized Lorentzian. The proof involves certain graphs on ; more strongly, we explain why the character (i.e., generating function) of the Kostant partition function of any loopless multigraph on is Lorentzian after shifting and normalizing. We then show that parabolic Vermas form a "maximal" class with log-concave (hence DL) characters. Namely, log-concavity fails in greater generality along three natural directions: (1) it does not hold for every simple Lie type, (2) nor for a larger universal family of highest weight modules, the higher order Verma modules, even in type , and (3) it does not always hold for important generalizations of Schur polynomials: the Jack and Macdonald polynomials. Finally, we extend these results to parabolic (i.e. "first order") and higher order Verma modules over the semisimple Lie algebras . We also partially resolve a conjecture of Huh et al on the DL property for integral highest weight simple modules.

Paper Structure

This paper contains 25 sections, 15 theorems, 119 equations, 1 table.

Key Result

Theorem 1.4

Suppose $h(x) = \sum_{\mu\in\mathbb{N}^m} c_\mu x^{\mu}$ is denormalized Lorentzian and nonzero. Then:

Theorems & Definitions (45)

  • Definition 1.1
  • Definition 1.2
  • Definition 1.3: BH
  • Theorem 1.4
  • Theorem 1.5
  • Theorem 1.6
  • Theorem 1.7
  • Theorem 1.8
  • Conjecture 1.9
  • Definition 3.1
  • ...and 35 more