Partial Springer Decomposition Overview

• Partial Springer decomposition is a framework that replaces full Springer fibers with canonical, tractable decompositions via restricted regions, Schubert models, or partial sheaves.
• It employs diverse methodologies including geometric restrictions, affine pavings, and homological techniques to preserve essential Springer structures and representation data.
• This approach offers actionable insights by categorizing complex Springer fibers through the isolation of positive cells, cohomological summands, and invariant modules.

Searching arXiv for recent and foundational papers related to “Partial Springer Decomposition”. “Partial Springer decomposition” is not a single uniformly standardized construction. In the literature represented here, it designates several related ways of extracting a tractable decomposition from Springer-theoretic geometry: by restricting to a distinguished region of a Springer fiber, by replacing a Springer fiber with a cohomologically equivalent Schubert model, or by passing from varieties to partial Springer resolutions, sheaves, or top homology. The common feature is that one does not decompose the full Springer fiber in the strongest classical sense, but instead obtains a canonical decomposition, stratification, or representation-theoretic shadow that retains part of the Springer structure (Lusztig, 2019, Precup et al., 2017, Dong et al., 2021).

1. General framework and scope

A broad formal setting is given by the survey framework in which one starts with a connected reductive group $G$, parabolics $P_i\subset G$, a $G$-representation $V$, and $P_i$-stable subrepresentations $F_i\subset V$. One then forms homogeneous bundles

$E_i=G\times^{P_i}F_i,$

their disjoint union $E=\bigsqcup_i E_i$, a Springer map $\pi:E\to V$, and the Steinberg variety

$Z=E\times_VE.$

In this language, classical Springer theory is the special case $P_i\subset G$0, while partial or parabolic variants arise by allowing proper parabolics. The equivariant BBD decomposition theorem is recorded in the form

$P_i\subset G$1

so partial Springer problems are naturally tied to IC-summands, graded multiplicity spaces, and modules over the associated Steinberg algebra rather than only to cells inside a fiber (Sauter, 2013).

A related generalized viewpoint appears in type $P_i\subset G$2, where an extended Springer-type map

$P_i\subset G$3

is used to realize all Lusztig sheaves at once. The direct image satisfies

$P_i\subset G$4

and each $P_i\subset G$5 further decomposes as

$P_i\subset G$6

This suggests that “partial Springer decomposition” can also mean a generalized or parabolic replacement of the ordinary Springer resolution by a map whose decomposition theorem recovers a larger package of orbit-local-system data (Graham et al., 2020).

2. Totally positive Springer fibers as a literal partial decomposition

The most explicit geometric realization of the phrase occurs in Lusztig’s study of total positivity in Springer fibers. Let $P_i\subset G$7 be unipotent, and define the positive part of the Springer fiber by

$P_i\subset G$8

The ambient positive flag manifold has Rietsch’s cell decomposition

$P_i\subset G$9

If $G$0, with

$G$1

then Theorem 1.14 identifies the positive Springer fiber by fixed-point conditions: $G$2 Corollary 1.16 then gives the decomposition

$G$3

Each piece is already a Rietsch cell, hence homeomorphic to $G$4, and the indexing depends only on the support data $G$5, not on the positive parameters of $G$6 inside $G$7 (Lusztig, 2019).

This construction is “partial” in two precise senses. First, it concerns only

$G$8

not the whole Springer fiber $G$9. Second, the cells are not newly defined Springer-theoretic strata intrinsic to $V$0; they are exactly those cells of the positive flag manifold that happen to lie in $V$1. The paper also gives a partial flag analogue: $V$2 In examples for $V$3 and $V$4, the resulting positive pieces are unions of zero-, one-, and two-dimensional cells and can be contractible without being pure-dimensional, which sharply distinguishes them from the full Springer fiber (Lusztig, 2019).

3. Schubert models and cohomological partial decompositions

A different sense of partial decomposition appears in the work of Precup and Tymoczko on type $V$5 Springer fibers. For $V$6 and a nilpotent $V$7 of Jordan type $V$8, the Springer fiber $V$9 has an affine paving by

$P_i$0

where $P_i$1 is a Schubert cell. To a row-strict tableau $P_i$2 one associates a Schubert point $P_i$3 satisfying

$P_i$4

When $P_i$5 has at most three rows or at most two columns, the main theorem gives the Poincaré-polynomial identity

$P_i$6

The key intermediate statement is that the Schubert points form a Bruhat lower order ideal in these shape classes: if $P_i$7 is a Schubert point and $P_i$8, then $P_i$9 is also a Schubert point associated to $F_i\subset V$0 (Precup et al., 2017).

This is not a geometric decomposition of $F_i\subset V$1 into Schubert varieties. The paper explicitly rules out an isomorphism

$F_i\subset V$2

and equally rules out a literal decomposition of $F_i\subset V$3 by Schubert varieties. What survives is a dimension-preserving bijection between the affine paving cells of the Springer fiber and the Schubert cells in a specific union of Schubert varieties. The outcome is therefore a cohomological or Betti-number decomposition model. Its scope is also sharply limited: it is proved only for partitions with at most three rows or at most two columns, fails in arbitrary Lie type, and the key deletion lemma already breaks for shapes containing $F_i\subset V$4 as a subdiagram (Precup et al., 2017).

4. Partial Springer fibers, top homology, and partial Springer sheaves

In type $F_i\subset V$5, partial Springer decomposition appears at the level of top Borel–Moore homology rather than cells. For partial isotropic flag varieties $F_i\subset V$6, the partial Springer resolution

$F_i\subset V$7

has fibers $F_i\subset V$8, and the union over $F_i\subset V$9 gives the partial Springer fiber $E_i=G\times^{P_i}F_i,$0. The main theorem identifies the top homology as a $E_i=G\times^{P_i}F_i,$1-module: $E_i=G\times^{P_i}F_i,$2 Here $E_i=G\times^{P_i}F_i,$3 is the type $E_i=G\times^{P_i}F_i,$4 Weyl group, $E_i=G\times^{P_i}F_i,$5 ranges over bipartitions, $E_i=G\times^{P_i}F_i,$6 is the type $E_i=G\times^{P_i}F_i,$7 nilpotent partition attached by Springer correspondence, and $E_i=G\times^{P_i}F_i,$8 is the irreducible module obtained from Schur–Weyl duality. Thus the decomposition is genuine at the sheaf and homology level, but not a stratification of the partial Springer fiber as a variety (Dong et al., 2021).

A complementary sheaf-theoretic formulation is given by the theory of partial Springer sheaves. For a parabolic $E_i=G\times^{P_i}F_i,$9 with Levi $E=\bigsqcup_i E_i$0, one has partial Grothendieck and Springer sheaves $E=\bigsqcup_i E_i$1 and $E=\bigsqcup_i E_i$2. Their endomorphism algebra is

$E=\bigsqcup_i E_i$3

where

$E=\bigsqcup_i E_i$4

is the relative Weyl group, and

$E=\bigsqcup_i E_i$5

The comparison between the restriction construction and the Fourier-transform construction of the $E=\bigsqcup_i E_i$6-action takes the form

$E=\bigsqcup_i E_i$7

where $E=\bigsqcup_i E_i$8 and $E=\bigsqcup_i E_i$9 is the one-dimensional character by which $\pi:E\to V$0 acts on $\pi:E\to V$1. Since $\pi:E\to V$2 need not be a Coxeter group, the classical sign character is replaced by this top-cohomology character. This makes clear that in partial Springer theory, decomposition questions often move from cells in fibers to endomorphism algebras, invariant parts, and relative Weyl-group symmetries (Chatterjee et al., 2024).

5. Delta-Springer varieties and generalized partial-flag realizations

$\pi:E\to V$3-Springer varieties provide a concrete family of generalized Springer fibers inside partial flag varieties. For a partition $\pi:E\to V$4 and $\pi:E\to V$5, they are defined by

$\pi:E\to V$6

In the two-row case $\pi:E\to V$7, this becomes a subvariety of

$\pi:E\to V$8

with $\pi:E\to V$9 recovering the ordinary two-row Springer fiber and $Z=E\times_VE.$0 recovering an exotic Springer fiber. The irreducible components are indexed by $Z=E\times_VE.$1-cup diagrams $Z=E\times_VE.$2; for each such diagram, the corresponding component $Z=E\times_VE.$3 is a $Z=E\times_VE.$4-fold iterated fiber bundle over $Z=E\times_VE.$5, hence smooth, and

$Z=E\times_VE.$6

is a bijection from $Z=E\times_VE.$7 to the irreducible components of $Z=E\times_VE.$8. The same varieties also admit an affine paving indexed by $Z=E\times_VE.$9-weights (Lacabanne et al., 2024).

A related generalized Springer interpretation is given by the varieties $P_i\subset G$00, whose cohomology modules are $P_i\subset G$01-Springer modules. These are proved to be Borho–MacPherson generalized Springer fibers: $P_i\subset G$02 If

$P_i\subset G$03

then their graded Frobenius series satisfies the skewing formula

$P_i\subset G$04

Thus the cohomology of the generalized fiber is obtained from ordinary Springer cohomology by extracting a Levi-isotypic piece, expressed symmetrically by the skewing operator $P_i\subset G$05. This is a particularly transparent example in which “partial Springer decomposition” means controlled passage from a full Springer module to a partial-flag or generalized Springer module (Gillespie et al., 2023).

6. Modular, super, and affine partiality

In modular Springer theory, partial decomposition arises as a replacement for a failed decomposition theorem. For the Springer resolution $P_i\subset G$06 over a field $P_i\subset G$07, Shchigolev studies the perverse sheaf

$P_i\subset G$08

when semisimplicity fails. For an orbit representative $P_i\subset G$09, component group $P_i\subset G$10, and irreducible $P_i\subset G$11-module $P_i\subset G$12, the paper proves

$P_i\subset G$13

Semisimplicity holds if and only if all radicals vanish and the top cohomology modules $P_i\subset G$14 are semisimple as $P_i\subset G$15-modules. Here partial Springer decomposition means orbitwise control of composition factors rather than an actual direct-sum splitting (Shchigolev, 2014).

A different nonclassical form occurs for the odd nilcone of $P_i\subset G$16. For $P_i\subset G$17 in the odd nilcone, the Springer fiber $P_i\subset G$18 admits a decomposition

$P_i\subset G$19

into locally closed subsets indexed by admissible slicings of the nilpotent orbit diagram $P_i\subset G$20. This decomposition is recursive and canonical, but unlike the classical case the strata need not be equidimensional, their closures need not be irreducible components, and fibers can be disconnected. The result is therefore a genuine partial stratification rather than a component decomposition (Leidwanger et al., 2010).

In the affine setting for $P_i\subset G$21, Chen proves a decomposition theorem for finite abelian covers

$P_i\subset G$22

deforming finite coverings of compactified Jacobians, decomposing

$P_i\subset G$23

into intersection complexes. Passing to the limit yields a decomposition formula for

$P_i\subset G$24

the homology of the affine Springer fiber, in which the main term computes

$P_i\subset G$25

while the remaining terms are Levi-type contributions built recursively from smaller spectral pieces. In this context the “partial” feature is the reduction of full affine Springer homology to its $P_i\subset G$26-invariant subspace together with explicitly controlled lower-rank correction terms (Chen, 2024).

Taken together, these constructions show that partial Springer decomposition is best understood as a family of restricted, relative, or replacement decompositions. What is preserved varies—from positive cells, to Schubert models, to top homology summands, to IC constituents—but the guiding principle is constant: one replaces the full complexity of Springer fibers or Springer sheaves by a decomposition that is canonical on a smaller region, in a weaker category, or after passage to an associated representation-theoretic shadow.