---
title: Affine Brylinski Filtration in Kac–Moody Algebras
url: https://www.emergentmind.com/topics/affine-brylinski-filtration
type: topic
---

# Affine Brylinski Filtration in Kac–Moody Algebras

The affine Brylinski filtration is the affine analogue of the Brylinski–Kostant filtration on representations of finite-dimensional semisimple Lie algebras. In affine Kac–Moody theory, the finite-type principal nilpotent is replaced by the principal Heisenberg subalgebra, and the resulting filtration on dominant weight spaces of integrable highest-weight modules has associated graded Hilbert–Poincaré series equal to Lusztig’s $t$-analogue of weight multiplicities. For the basic representation $L(\Lambda_0)$, this filtration admits an explicit Poincaré–Birkhoff–Witt type basis; in simply-laced untwisted affine type, a uniform proof is obtained through $\mathscr W$-algebras, Drinfeld–Sokolov reduction, and the Miura map [1012.2095, 2508.10365]. A distinct arithmetic-geometric usage extends Brylinski’s filtration on Witt vectors to the de Rham–Witt complex [2601.12177].

## 1. Affine Kac–Moody setting and principal Heisenberg structure

Let $\bar{\mathfrak g}$ be a finite-dimensional simple Lie algebra of type $A_\ell$, $D_\ell$, or $E_6,E_7,E_8$, with Cartan subalgebra $\bar{\mathfrak h}$, root lattice $Q$, normalized invariant form $(\cdot\mid\cdot)$, and fundamental degrees
$$
d_1\le d_2\le \cdots \le d_\ell.
$$
Its untwisted affine Kac–Moody algebra is
$$
\mathfrak g=\bar{\mathfrak g}\otimes\mathbb C[t,t^{-1}] \oplus \mathbb C K \oplus \mathbb C d.
$$
If $e_i,f_i$ are the Chevalley generators, the principal Heisenberg subalgebra is
$$
\mathfrak s=\{x\in\mathfrak g\mid [x,e_0+\cdots+e_\ell]\in \mathbb C\cdot K\},
$$
with positive part $\mathfrak s^+=\mathfrak s\cap\mathfrak n^+$ [2508.10365].

In Slofstra’s formulation for untwisted affine Kac–Moody algebras, one starts with a simple finite-dimensional Lie algebra $\mathfrak g$ and its affine extension $\widehat{\mathfrak g}$, forms the principal nilpotent $e=\sum_{i=1}^{\ell} e_i$, and writes $e(n)=e\otimes t^n$. The span of $K$ and the $e(n)$ is then a Heisenberg algebra, and the positive part $\mathfrak s_{>0}=\bigoplus_{n>0}\mathbb C\,e(-n)$ is abelian and graded by $n>0$ [1012.2095]. This replacement of the principal nilpotent by the principal Heisenberg is the defining affine modification.

The basic representation is the level-$1$ vacuum module $L(\Lambda_0)$, on which $K$ acts by $1$. Its $\bar{\mathfrak h}\oplus \mathbb CK$-invariant subspace is
$$
Z=L(\Lambda_0)^{\bar{\mathfrak h}\oplus\mathbb CK}
   =\bigoplus_{n\ge 0} Z_n,
\qquad
Z_n=L(\Lambda_0)_{\Lambda_0-n\delta}.
$$
This $\delta$-string is the canonical arena for the explicit basis results proved via $\mathscr W$-algebras [2508.10365].

## 2. Definition of the affine Brylinski filtration

For an integrable highest-weight module $L(\Lambda)$ of positive level and a weight space $L(\Lambda)_\mu$, the affine Brylinski filtration may be written in two equivalent ways. In the Heisenberg-span form,
$$
F^nL(\Lambda)_\mu
=
\mathrm{Span}\Bigl\{
e(-m_1)\cdots e(-m_k)\,v_\Lambda
\ \Big|\
k\le n,\ m_j>0,\ \text{total weight } \mu
\Bigr\}.
$$
Because the positive Heisenberg part is abelian, this is equivalent to the annihilation formulation
$$
F^nL(\Lambda)
=
\{v\in L(\Lambda)\mid x_1x_2\cdots x_{n+1}\cdot v=0
\text{ for all }x_i\in \mathfrak s_+\},
$$
with $F^{-1}=\{0\}$ and $F^nL(\Lambda)_\mu=F^nL(\Lambda)\cap L(\Lambda)_\mu$ [1012.2095, 1912.13353].

For the basic representation in simply-laced affine type, the filtration is written as
$$
F^iL(\Lambda_0)
=
\{v\in L(\Lambda_0): x^{\,i+1}\cdot v=0 \ \forall\,x\in s^+\},
\qquad i\ge -1.
$$
Restricting to the invariant subspace gives
$$
F^iZ_n=Z_n\cap F^iL(\Lambda_0),
$$
and hence a bi-graded associated graded space
$$
\operatorname{gr} Z
=
\bigoplus_{n,i\ge 0} F^iZ_n/F^{i-1}Z_n.
$$
The filtration degree records Heisenberg complexity, while the $\delta$-degree records the position along the $\delta$-string [2508.10365].

In the principal vertex-operator realization discussed in the type $A$ case, the positive Heisenberg modes act compatibly with the filtration: for $h\in\mathfrak h$ and $j>0$ one has
$$
h(j)^{(o)}\cdot F^nL(\Lambda_0)\subset F^{n-1}L(\Lambda_0),
$$
while for $j<0$,
$$
h(j)^{(o)}\cdot F^nL(\Lambda_0)\subset F^{n+1}L(\Lambda_0).
$$
This realizes the filtration as an operator-theoretic grading in the principal model [1912.13353].

## 3. Hilbert–Poincaré series and Lusztig’s $t$-analogue

The principal structural theorem is that the affine Brylinski filtration recovers Lusztig’s $t$-analogue of weight multiplicity on dominant weights. If $\Lambda$ is dominant integral of positive level and $\mu$ is dominant, then the Poincaré series
$$
P_{L(\Lambda),\mu}(q)
=
\sum_{n\ge 0}
\dim\bigl(F^nL(\Lambda)_\mu/F^{n-1}L(\Lambda)_\mu\bigr)\,q^n
$$
coincides with Lusztig’s polynomial
$$
m^\Lambda_\mu(q)
=
\sum_{w\in W}(-1)^{\ell(w)}\,K(w\cdot\Lambda-\mu;q).
$$
Equivalently, the Hilbert–Poincaré series of the associated graded dominant weight space is exactly Lusztig’s $t$-analogue [1012.2095, 1912.13353].

For the basic representation, the invariant subspace $Z$ admits a closed two-variable Hilbert series. With $d_k$ the fundamental degrees of $\bar{\mathfrak g}$,
$$
\mathrm{Hilb}(\operatorname{gr} Z;t,q)
:=
\sum_{n,i\ge 0}
\dim(F^iZ_n/F^{i-1}Z_n)\,t^i q^n
=
\prod_{k=1}^{\ell}\prod_{m=1}^{\infty}(1-t^{d_k}q^m)^{-1}.
$$
This formula is attributed in the paper to GSV and Slofstra, and in the type $A$ case its double-product form is identified with the $q$-Macdonald–Mehta constant-term identity [2508.10365, 1912.13353].

The significance of this equality is that the filtration is not merely an internal module-theoretic construction. It computes a canonical graded multiplicity polynomial already appearing in affine representation theory and in conjectural affine Satake-type frameworks. Slofstra’s paper states this as the affine replacement for the finite-dimensional Brylinski–Kostant picture proposed in relation to Braverman–Finkelberg’s conjectural analogue of the geometric Satake isomorphism [1012.2095].

## 4. $\mathscr W$-algebra realization and explicit PBW basis

The $\mathscr W$-algebra description begins with the lattice vertex algebra
$$
V_Q=\pi_1\otimes \mathbb C[Q]_\epsilon,
$$
where $\pi_1$ is the level-$1$ Heisenberg Fock space of $\bar{\mathfrak h}\otimes\mathbb C[t,t^{-1}]$ and $\epsilon$ is a standard $2$-cocycle. The $\mathscr W$-algebra of $\bar{\mathfrak g}$ is
$$
W=\bigcap_{x\in\bar{\mathfrak g}}\Ker_{V_Q} x_{(0)}.
$$
By Feigin–Frenkel, it is freely generated as a vertex algebra by fields
$$
\omega^{(d_1)},\ldots,\omega^{(d_\ell)}
$$
of conformal weights $d_1,\ldots,d_\ell$ [2508.10365].

Fix the principal twisted realization $M_\sigma\simeq L(\Lambda_0)$ of $V_Q$ and restrict it to $W$. In the type $A$ formulation, the dominant weight spaces together form an irreducible Verma module of the corresponding $W$-algebra, and the natural PBW basis of this module is compatible with the Brylinski filtration [1912.13353]. The simply-laced uniform theorem states that for each $n,d\ge 0$, the subspace $F^dZ_n$ has a basis consisting of vectors
$$
\omega^{(p_1)}_{k_1}(\sigma)\,
\omega^{(p_2)}_{k_2}(\sigma)\cdots
\omega^{(p_r)}_{k_r}(\sigma)\cdot v_0
$$
subject to the conditions
\[
\begin{aligned}
&\text{(1)}\quad r\ge 0,\ \ \ell\ge p_1\ge p_2\ge\cdots\ge p_r\ge 1,\\
&\text{(2)}\quad k_j\le -1 \text{ for all }j,\ \ \sum_j k_j=-n,\\
&\text{(3)}\quad p_i=p_{i+1}\Rightarrow k_i\le k_{i+1},\\
&\text{(4)}\quad \sum_j d_{p_j}\le d.
\end{aligned}
\]
Here $v_0$ is the highest-weight vacuum in $M_\sigma$ [2508.10365].

These monomials realize the two gradings explicitly: the $t$-degree is $\sum_j d_{p_j}$ and the $q$-degree is $n$. In the type $A$ paper, the analogous statement is expressed by saying that PBW monomials of total degree at most $n$ span $F^nZ$, and that modes $w_p(k)$ with $k\le -1$ raise the filtration index by exactly $d_p$ [1912.13353]. The resulting basis is the analogue, for the principal vertex-operator realization of $L(\Lambda_0)$, of Feigin–Frenkel’s basis of $\mathcal W$.

## 5. Uniform simply-laced proof and the failure of the principal nilpotent

The type-free proof for simply-laced affine Lie algebras proceeds by identifying $Z$ inside the principal realization with a $\mathscr W$-module and then proving that it is an irreducible $\mathscr W$-Verma module. More precisely, the paper identifies $Z$ with the space of invariants under the positive Heisenberg and shows that, as a $W$-module,
$$
Z\simeq M^W(\bar\gamma_{h^\vee-\rho}),
$$
an irreducible Verma module of $W$ at central charge
$$
c=\operatorname{rank}\bar{\mathfrak g}.
$$
It also states, as a corollary, that $Z$ is a $W$-Verma module at level $k+1=1$ and highest weight $\mu=\rho-h^\vee$ [2508.10365].

The proof uses Drinfeld–Sokolov reduction in the form
$$
H^0_{DS}(M(\lambda))\simeq M^W(\bar\gamma_\lambda)
$$
for generic affine weight $\lambda$, together with the Kac–Kazhdan criterion to deduce irreducibility of the $W$-Verma from irreducibility of the affine Verma. It then invokes the Miura map
$$
\Upsilon: W_k(\bar{\mathfrak g})\hookrightarrow \pi_{k+h^\vee},
\qquad k\neq -h^\vee,
$$
which at $k+1=1$ identifies $W\simeq \operatorname{image}(\Upsilon)\subset \pi_1$. For $\mu=\rho-h^\vee$, the free-field PBW basis of the highest-weight Fock module $\pi_{1,\mu}$ pulls back to the $\omega$-monomial basis of the filtration subspaces [2508.10365].

A recurrent misconception is that the finite-type principal nilpotent filtration should extend verbatim to affine type. Slofstra gives an explicit counterexample in $\widehat{\mathfrak{sl}}_2$. Writing the principal nilpotent as
$$
e=E+F\,t
$$
and taking
$$
w=(F\,t^{-1})(E\,t^{-1})\,v
$$
in the level-$1$ module $L(\Lambda_1)$, one has
$$
e^2w=0\quad\text{but}\quad ew\neq 0.
$$
The corresponding Poincaré series differ:
$$
P_{e\text{-filtration}}(q)=q+q^4,
\qquad
P_{\text{Heisenberg-filtration}}(q)=q^2+q^4
=
m^{\Lambda_1}_{(0,1,-2)}(q).
$$
This is the basic reason that the affine theory is formulated with the principal Heisenberg rather than the principal nilpotent [1012.2095].

## 6. Alternative arithmetic-geometric usage

A distinct construction, also described as an affine Brylinski filtration, appears in positive-characteristic arithmetic geometry. Let $(X,E)$ be an snc-pair over a perfect field of characteristic $p$, with
$$
E=\sum_{i=1}^r E_i,
\qquad
D=\sum_{i=1}^r n_iE_i,
\qquad
U=X\setminus E.
$$
Locally, if $\{x_i\}_{i=1}^r$ is a regular system of parameters defining the components of $E$ and
$$
\pi=\prod_{i=1}^r x_i,
$$
one defines
$$
{}_DW_m(A_\pi)
=
\Bigl\{
\underline a=(a_{m-1},\dots,a_0)\in W_m(A_\pi)
\ \Big|\
a_i^{p^i}\in \pi^{-n_i}A\ \forall i
\Bigr\},
$$
and then
$$
{}_DW_m\Omega^q_{X\setminus E}
=
{}_DW_m\Omega^q_X(\log E)
+
d\bigl({}_DW_m\Omega^{q-1}_X(\log E)\bigr)
\subset j_*W_m\Omega^q_U.
$$
This sheaf-theoretic filtration cuts out Witt-differential forms whose Witt-coordinate entries have poles of order $\le r$ along each component of $E$ [2601.12177].

When $X=\operatorname{Spec}A$ is a henselian DVR with fraction field $K$, this construction recovers Brylinski’s original filtration
$$
\Fil^rW_m(K)
=
\Bigl\{
\underline a\in W_m(K)\ \Big|\ a_i^{p^i}\pi^r\in A\ \forall i
\Bigr\},
$$
through the identity
$$
{}_{rE}W_m\Omega^0_U={}_{rE}W_m(K)=\Fil^rW_m(K).
$$
For arbitrary degree $q$, Krishna–Majumder give an explicit description of ${}_{rE}W_m\Omega^q_K$ in terms of generators of the form
$$
V^s\bigl(a[\pi]^j_{m-s}\bigr)
\quad\text{and}\quad
dV^s\bigl(b[\pi]^j_{m-s}\bigr),
$$
with $a\in W_{m-s}\Omega^q_k$ and $b\in W_{m-s}\Omega^{q-1}_k$ [2601.12177].

The same paper constructs the two-term complex
$$
W_m^{q,\bullet}\big|_{D_n}
=
\bigl[
Z_1\,{}_{D_n}W_m\Omega^q
\xrightarrow{\,1-C\,}
{}_{D_n}W_m\Omega^q
\bigr]
$$
and proves that its hypercohomology recovers Kato’s lower numbering subgroups in
$$
H^{q+1}(K)\{p\}=H^1(K,W_m\Omega^q)\quad (m\to\infty).
$$
It also proves that Frobenius, Verschiebung, and restriction preserve the filtration, satisfying
$$
F({}_DW)={}_DW,\qquad V({}_DW)={}_DW,\qquad R({}_DW)={}_{D/p}W.
$$
This usage is mathematically separate from the affine Kac–Moody filtration, but both constructions extend a Brylinski filtration from a degree-zero setting to a richer graded or sheaf-theoretic context [2601.12177].

Source: https://www.emergentmind.com/topics/affine-brylinski-filtration