Almost-Uniform Steiner Bundles

• Almost-uniform Steiner bundles are vector bundles on projective spaces defined by two-step linear resolutions with generic balanced splitting and a controlled jumping locus.
• They feature a constant splitting type on general lines, with deviations constrained to determinantal loci that capture exceptional behavior.
• Their study connects cohomological properties, slope stability, and a Kronecker-quiver formulation, offering constructive existence results and geometric insights.

Almost-uniform Steiner bundles are Steiner bundles whose restriction data are generically constant but admit a controlled jumping locus. On projective space, the basic model is a bundle $V$ on $\mathbb{P}^n$ presented by a general linear resolution

$$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$

and the almost-uniform phenomenon appears in the splitting of $V$ on lines: for general lines the splitting is balanced and constant, while deviations are confined to a determinantal locus of special lines or special matrices (Coskun et al., 2022). The term is not completely uniform across the literature: in the projective-space analysis of general Steiner bundles it is an inferred property rather than a formal definition, whereas in later work on $\mathbb{P}^n$, following Ellia, almost-uniformity means that the set of jumping lines is finite and non-empty (Bissinger, 2 Sep 2025).

1. Steiner resolutions and numerical invariants

A Steiner bundle on $\mathbb{P}^n$ in the sense of Dolgachev–Kapranov is a vector bundle $V$ admitting an exact sequence

$$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$

with $t>0$, $r\ge n$, and general map. Here “general” means a member of a dense open subset of the affine parameter space $\mathbb{P}^n$0, so that the sequence is exact and $\mathbb{P}^n$1 is locally free when $\mathbb{P}^n$2. The dual presentation

$\mathbb{P}^n$3

is equally fundamental, because the maps on global sections govern the cohomological behavior of twists $\mathbb{P}^n$4 (Coskun et al., 2022).

The basic invariants are determined directly by the resolution. Writing $\mathbb{P}^n$5 for the hyperplane class, one has

$\mathbb{P}^n$6

Hence

$\mathbb{P}^n$7

and, for $\mathbb{P}^n$8,

$\mathbb{P}^n$9

truncated at $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$0. These formulas are the numerical substrate for the later cohomology, stability, ampleness, and restriction statements. They also show that, within a fixed pair $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$1, the parameter space of general Steiner bundles is governed by the linear data of the defining matrix.

A first structural point is that almost-uniformity is not an extra decoration imposed on an arbitrary bundle class. For Steiner bundles it is encoded already at the level of the two-step linear resolution: the same ratio $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$2 that fixes slope and border cohomological twists also fixes the generic splitting profile on lines. This suggests that almost-uniformity is an intrinsic asymptotic consequence of the Steiner presentation rather than an accidental feature of isolated examples.

2. Restriction to lines and balanced splitting

Restricting the defining sequence to a line $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$3 gives

$$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$4

Therefore $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$5 and $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$6. For a general matrix and a general line, the restricted map is general, so $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$7 is a general quotient of $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$8 by $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$9. Since every vector bundle on $V$0 splits, the splitting on a general line is balanced and uniquely determined by $V$1 and $V$2: $V$3 This balanced splitting is constant on a dense open subset of the Grassmannian of lines, so the restriction type is uniform on general lines (Coskun et al., 2022).

Within this projective-space framework, the paper does not formally define “almost-uniform,” but its restriction analysis yields a precise almost-uniform property: the splitting type is constant and balanced on a dense open set of lines, and the jumping locus where the splitting deviates is a proper closed subset. The explicitly analyzed jump is the appearance of a trivial summand $V$4, equivalently $V$5. For a fixed line this condition cuts codimension at least $V$6 in the fiber $V$7, while globally the locus of maps admitting some such line has codimension at least $V$8 in the parameter space (Coskun et al., 2022).

A different, stricter convention appears in later work on $V$9: for a bundle $\mathbb{P}^n$0, there is a dense open subset $\mathbb{P}^n$1 of the Grassmannian of lines on which the splitting multiplicities are constant, and the complementary set $\mathbb{P}^n$2 is the set of jumping lines; following Ellia, $\mathbb{P}^n$3 is almost-uniform if $\mathbb{P}^n$4 is finite and non-empty. In this language, uniformity means $\mathbb{P}^n$5, while almost-uniformity means that the failure of uniformity is concentrated on finitely many lines (Bissinger, 2 Sep 2025).

A recurrent misconception is to identify almost-uniformity with homogeneity. The later representation-theoretic results show that these notions are distinct: uniform Steiner bundles can be non-homogeneous, and almost-uniformity concerns the geometry of the jumping locus rather than invariance under the action of $\mathbb{P}^n$6.

3. Cohomology, stability, and ampleness

The cohomological behavior of general Steiner bundles is organized by a natural cohomology conjecture: every twist $\mathbb{P}^n$7 should have at most one nonvanishing cohomology group. For the dual kernel presentation

$\mathbb{P}^n$8

set

$\mathbb{P}^n$9

Then $\mathbb{P}^n$0 has natural cohomology if and only if

$\mathbb{P}^n$1

This reduces all twists to two border cases via Castelnuovo–Mumford regularity and semicontinuity. The conjecture is proved on $\mathbb{P}^n$2, and asymptotically it holds under scaling: for

$\mathbb{P}^n$3

if

$\mathbb{P}^n$4

then $\mathbb{P}^n$5 has natural cohomology. There is also an exact divisibility result: if $\mathbb{P}^n$6, then the general Steiner bundle on $\mathbb{P}^n$7 has natural cohomology (Coskun et al., 2022).

Stability is governed by a concrete slope window. If

$\mathbb{P}^n$8

then a general Steiner bundle

$\mathbb{P}^n$9

is slope stable whenever

$V$0

For degree-$V$1 maps

$V$2

the same numerical range yields slope semistability for general $V$3. In $V$4, the low-slope regime is handled by restriction to a quadric and the use of Abe’s and Rudakov’s classification (Coskun et al., 2022).

Ampleness connects directly with the line-splitting picture. There is a closed subset $V$5 in the map space of codimension at least $V$6 such that if the defining matrix lies outside $V$7, then $V$8 is ample. In particular, if

$V$9

the general Steiner bundle is ample. Since ampleness on lines is equivalent to positivity of all summands of $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$0, this criterion turns the almost-uniform statement into a global positivity statement: in the ample range, every line carries only positive summands; below that range, failure of ampleness is still confined to a determinantal bad locus of controlled codimension (Coskun et al., 2022).

The cohomology, stability, and restriction numerics are aligned by the same ratio $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$1. The border twists $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$2, the balanced degrees on general lines, and the slope window are all controlled by this single parameter. A plausible implication is that almost-uniformity for general Steiner bundles is not merely geometric but also cohomological: the same numerics constrain both line restrictions and the placement of nonzero cohomology.

4. Kronecker-quiver formulation

A decisive reorganization of Steiner bundles on $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$3 comes from the generalized Kronecker quiver $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$4. In the notation of the representation-theoretic literature, one sets $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$5 and identifies $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$6. A bundle $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$7 is Steiner if there are finite-dimensional $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$8-vector spaces $$0 \to \mathcal{O}_{\mathbb{P}^n}(-1)^t \to \mathcal{O}_{\mathbb{P}^n}^{t+r} \to V \to 0,$$9 and an exact sequence

$t>0$0

There is an equivalence

$t>0$1

from relative $t>0$2-projective Kronecker representations to Steiner bundles. If the corresponding representation $t>0$3 has dimension vector $t>0$4, then

$t>0$5

Under this equivalence, the restriction of $t>0$6 to a line is read off from the decomposition of the restricted representation $t>0$7 into preprojective indecomposables $t>0$8, and the multiplicities of $t>0$9 are exactly the splitting multiplicities of $r\ge n$0 on that line (Bissinger, 2 Sep 2025).

This dictionary introduces the support

$r\ge n$1

and the $r\ge n$2-type

$r\ge n$3

which records the highest twisting degree appearing in the splitting type on lines. It also separates homogeneity from uniformity. A bundle is homogeneous if $r\ge n$4 for every $r\ge n$5, and homogeneous bundles are uniform; however, the converse fails.

The mechanism controlling splitting types is categorical. Restriction, inflation, and reflection functors produce adjoint pairs

$r\ge n$6

and from preprojective $r\ge n$7-modules one obtains test representations $r\ge n$8 attached to lines $r\ge n$9. Their Hom-vanishing detects whether the restricted representation is supported in degrees $\mathbb{P}^n$00 and $\mathbb{P}^n$01: $\mathbb{P}^n$02 Thus the line-splitting problem becomes a problem of orthogonality to test modules (Bissinger, 2 Sep 2025).

5. Existence results and prescribed jumping behavior

The representation-theoretic approach produces explicit existence theorems beyond the classical homogeneous examples. For any $\mathbb{P}^n$03 and $\mathbb{P}^n$04, there exists a simple uniform Steiner bundle on $\mathbb{P}^n$05 of $\mathbb{P}^n$06-type with support $\mathbb{P}^n$07 that is not homogeneous. There also exists an indecomposable uniform Steiner bundle on $\mathbb{P}^n$08 of $\mathbb{P}^n$09-type with support $\mathbb{P}^n$10 that is not homogeneous. Since $\mathbb{P}^n$11 is disconnected exactly for $\mathbb{P}^n$12, this yields uniform but non-homogeneous Steiner bundles with disconnected splitting type for every $\mathbb{P}^n$13 (Bissinger, 2 Sep 2025).

A more quantitative existence statement fixes rank and first Chern class. Given $\mathbb{P}^n$14, $\mathbb{P}^n$15, and $\mathbb{P}^n$16, for each

$\mathbb{P}^n$17

the general Steiner bundle on $\mathbb{P}^n$18 of rank $\mathbb{P}^n$19 and first Chern class $\mathbb{P}^n$20 is simple, uniform of $\mathbb{P}^n$21-type with support $\mathbb{P}^n$22, and is not homogeneous. These results show that uniformity does not force homogeneous symmetry, even inside the rigid class of Steiner bundles.

Almost-uniform bundles in the finite-jumping-line sense are also constructed systematically. For any finite non-empty set of lines $\mathbb{P}^n$23, there is a full subcategory of Steiner bundles consisting of almost-uniform bundles whose set of jumping lines is exactly $\mathbb{P}^n$24, and this subcategory corresponds via $\mathbb{P}^n$25 to a wild subcategory of $\mathbb{P}^n$26. A concrete criterion is

$\mathbb{P}^n$27

If this set is finite and non-empty, then $\mathbb{P}^n$28 is almost-uniform, and outside the jumping lines the generic splitting type is

$\mathbb{P}^n$29

In particular, for any line $\mathbb{P}^n$30, the bundle

$\mathbb{P}^n$31

has rank $\mathbb{P}^n$32 and $\mathbb{P}^n$33 (Bissinger, 2 Sep 2025).

The wildness statement is conceptually important. It shows that once the jumping set is allowed to be finite and prescribed, almost-uniform Steiner bundles form a class with rich internal complexity rather than a near-classifiable boundary case between uniform and non-uniform behavior.

6. Grassmannians, jumping pairs, and broader generalizations

On Grassmannians, Steiner bundles are defined by the universal subbundle rather than by $\mathbb{P}^n$34. If $\mathbb{P}^n$35, an $\mathbb{P}^n$36-Steiner bundle $\mathbb{P}^n$37 is given by

$\mathbb{P}^n$38

with

$\mathbb{P}^n$39

There is a lower bound

$\mathbb{P}^n$40

The central replacement for line-splitting is the geometry of jumping pairs. If $\mathbb{P}^n$41 is the linear map encoding the Steiner resolution, then a pair $\mathbb{P}^n$42 is a jumping pair when $\mathbb{P}^n$43, and the jumping-pair variety satisfies

$\mathbb{P}^n$44

where $\mathbb{P}^n$45 is the generalized Segre embedding and $\mathbb{P}^n$46. The expected dimension has a lower bound, tangent-space calculations give an upper bound, and Steiner bundles whose jumping locus has maximal dimension are exactly Schwarzenberger bundles (Arrondo et al., 2012).

A parallel theory exists on arbitrary smooth projective varieties. For a strongly exceptional pair $\mathbb{P}^n$47, an $\mathbb{P}^n$48-Steiner bundle has resolution

$\mathbb{P}^n$49

In the special case $\mathbb{P}^n$50, the paper introduces jumping pairs

$\mathbb{P}^n$51

proves an expected lower bound and a tangent-space upper bound for its dimension, and classifies the extremal maximal-dimension case as Schwarzenberger. The paper does not explicitly define almost-uniformity, but its jumping-pair formalism makes “small jumping locus” the natural measure of departure from uniformity: empty jumping locus corresponds to uniformity, while a small proper jumping locus corresponds to generic constancy with controlled exceptional behavior (Arrondo et al., 2013).

The representation-theoretic Grassmannian formulation sharpens this viewpoint further. For $\mathbb{P}^n$52, there is an equivalence

$\mathbb{P}^n$53

and for $\mathbb{P}^n$54 the closed $\mathbb{P}^n$55-th rank variety

$\mathbb{P}^n$56

detects failure of uniformity. In this framing, uniformity at level $\mathbb{P}^n$57 is equivalent to $\mathbb{P}^n$58, and a natural working notion of almost-uniformity at level $\mathbb{P}^n$59 is that $\mathbb{P}^n$60 be a proper closed subset, so that the bundle is uniform on the dense open complement and jumps along $\mathbb{P}^n$61. The same jumping locus is detected by Hom-orthogonality to elementary test modules $\mathbb{P}^n$62: $\mathbb{P}^n$63 This gives a direct representation-theoretic model for controlled generic uniformity on Grassmannians (Bissinger et al., 2024).

Taken together, these developments place almost-uniform Steiner bundles between two extremes. At one end are uniform bundles, with constant restriction data everywhere; at the other are extremal Schwarzenberger bundles, whose jumping geometry is as large as the theory permits. Almost-uniform bundles occupy the intermediate regime where the generic restriction type is rigid but the exceptional set is still geometrically meaningful, whether as a high-codimension determinantal locus on $\mathbb{P}^n$64, a finite set of jumping lines, or a small rank variety in the Grassmannian and Kronecker settings.