The Self-Projecting Grassmannian

Basic Information

• Paper ID: 2511.21442
• Title: The Self-Projecting Grassmannian
• Authors: Alheydis Geiger (Max Planck Institute for Mathematics in the Sciences), Francesca Zaffalon (Max Planck Institute & Weizmann Institute)
• Classification: math.AG (Algebraic Geometry), math.CO (Combinatorics)
• Submission Date: November 26, 2025
• Paper Link: https://arxiv.org/abs/2511.21442

Abstract

This paper introduces the self-projecting Grassmannian (SGr(k,n)), an irreducible subvariety of the Grassmannian that parametrizes linear subspaces satisfying a generalized self-duality condition. The authors study its relationships with classical moduli spaces (such as moduli spaces of genus g curves with marked points) and other natural subvarieties of the Grassmannian. The self-projecting condition is further translated into the combinatorial language of matroids, introducing self-projecting matroids, and computational methods are employed to study their realization spaces within the self-projecting Grassmannian.

Research Background and Motivation

Problems to be Addressed

1. Generalizing Self-Duality: Classical self-dual Grassmannians SGr(k,2k) study point configurations satisfying Gale duality. This paper extends this concept to the case n≠2k, defining the self-projecting Grassmannian SGr(k,n).
2. Connecting Geometry and Combinatorics: Establishing a bridge between moduli spaces in algebraic geometry and combinatorial matroid theory, particularly investigating which matroids can be realized by self-projecting point configurations.
3. Characterizing Moduli Spaces: Exploring birational equivalences between the self-projecting Grassmannian and curve moduli spaces (such as M_{1,10} and M_{5,13}).

Significance

• Theoretical Importance: The self-projecting condition uniformly describes all isotropic vector subspaces relative to some non-degenerate diagonal bilinear form, providing new perspectives on orthogonal Grassmannians.
• Applied Value: Orthogonal Grassmannians and their totally positive parts have important applications in physics (scattering amplitudes in ABJM theory), the Ising model, and cosmology.
• Combinatorial Significance: The introduction of self-projecting matroids provides new research objects for matroid realization space theory.

Limitations of Existing Methods

• Classical self-duality theory applies only to the case n=2k
• For general n>2k, a unified geometric and combinatorial framework is lacking
• Research on matroid realization spaces has focused mainly on the general case, with insufficient study of special subvarieties (such as self-dual and self-projecting)

Core Contributions

1. Defining the Self-Projecting Grassmannian: Introduces SGr(k,n) as the Zariski closure of k-dimensional subspaces satisfying the condition X·Λ·X^t=0 (where Λ is a diagonal full-rank matrix), and proves its irreducibility (Theorem 2.10).
2. Dimension Formula: When 2k≤n≤(k+1 choose 2), proves that dim(SGr(k,n)) = k(n-k) - (k+1 choose 2) + n - 1.
3. Birational Equivalence with Moduli Spaces:
   • Proves M_{1,10} is birationally equivalent to X(4,9)^{sd} (Theorem 3.1)
   • Proves M_{5,13} is birationally equivalent to X(5,13)^{sd} (Theorem 3.5)
4. Introducing Self-Projecting Matroids: Defines self-projecting matroids as matroids without half-coloops, which is a natural generalization of self-dual matroids.
5. Computational Study of Realization Spaces:
   • For all rank-2 self-projecting matroids, proves S(M)=R(M) (Theorem 4.9)
   • For rank-3 matroids on at most 8 elements, all except U_{3,6} satisfy S(M)=R(M) (Theorem 4.10)
   • For rank-4 matroids on 9 elements, among 7181 realizable self-projecting matroids, at least 174 satisfy S(M)=R(M)≠∅, and at least 5400 satisfy S(M)⊊R(M) (Theorem 4.11)
6. FAIR Data Practices: Adopts the new .mrdi file format to store computational results, with code to be integrated into the OSCAR system and data stored in the oscarDB database.

Detailed Methodology

Task Definition

Input: k-dimensional vector space V⊆K^n (or k×n matrix X)
Output: Determine whether V is self-projecting, i.e., whether there exists λ∈(K*)^n such that X·diag(λ)·X^t=0
Constraint: 2k≤n (otherwise the self-projecting space is empty)

Core Definitions and Constructions

1. Equivalent Characterization of Self-Projecting Condition

Let V∈Gr(k,n) be represented by a k×n matrix M_V, and let ν:K^ℓ→K^{(ℓ+1 choose 2)} denote the second Veronese embedding. Define the multi-Veronese matrix ν(M_V) as the (k+1 choose 2)×n matrix obtained by applying ν to each column of M_V.

Key Equivalence: V is self-projecting ⟺ there exists λ∈(K*)^n such that ν(M_V)·λ=0

Advantages of this characterization:

• Linearizes the quadratic condition X·Λ·X^t=0
• Corresponds to the intersection with the quadratic surface of self-dual forms
• Facilitates ideal definition and computation

2. Ideal Characterization

In dual Stiefel coordinates: $I_k,n^{sd} = (\langle ν(X)·λ \rangle : \langle λ_1···λ_n \rangle^∞) ∩ K[x_{(1,1)},...,x_{(k,n)}]$

In dual Plücker coordinates: $J_k,n^{sd} := (\langle ν(D_{k,n})·λ \rangle : \langle λ_1···λ_n \rangle^∞) ∩ K[q_I | I∈\binom{[n]}{k}] + I_{k,n}$

where D_{k,n} is the cocircuit matrix, defined as: $(D_{k,n})_{I,j} = \begin{cases} \text{sign}(I,j)q_{I∪j} & \text{if } j∉I \\ 0 & \text{else} \end{cases}$

3. Relationship with Orthogonal Grassmannians

Given λ∈(K*)^n, the orthogonal Grassmannian OGr^λ(k,n) is the set of V∈Gr(k,n) satisfying ω(u,v)=0 for all u,v∈V, where ω is defined by the diagonal matrix diag(λ).

Key Theorem 2.9:

• If K is algebraically closed, SGr_K(k,n) is the Zariski closure of the image of OGr_K^1(k,n) under the (K*)^n torus action
• If K=ℝ, SGr_ℝ(k,n) is the Zariski closure of the image of ∪{u∈{±1}^n} OGr_ℝ^u(k,n) under the (ℝ{>0})^n action

This shows that the self-projecting Grassmannian is the closure of the union of all diagonal orthogonal Grassmannians.

Technical Innovations

1. Application of Multi-Veronese Matrices

Linearizing the quadratic condition through ν(X) is a key technical innovation:

• Enables the use of linear algebra methods to study quadratic constraints
• The rank condition rank(ν(X))≤n-1 is equivalent to self-projectivity
• Establishes connection with the parameter space X^{r,d}_{m,n} of Caminata-Moon-Schaffler

2. Half-Coloop Concept

Definition: An element e∈n is a half-coloop if there exist rank-(k-1) flats F_1, F_2 such that F_1∪F_2=n{e}.

This concept:

• Generalizes the notion of coloop
• Reduces to self-duality when n=2k
• Provides a computable combinatorial criterion for n>2k

3. Computational Algorithm for Realization Spaces

Algorithms 1 and 2 compute realization spaces via Gröbner bases:

• Utilize matroid basis characterizations as constraints
• Saturate with respect to products of λ_i
• Eliminate to obtain pure ideals in matrix element variables
• Work in localized rings to ensure non-degeneracy

Optimization strategy: Select isomorphic matroids containing frames (i.e., size-(k+1) circuits containing basis k), significantly improving computational efficiency.

Experimental Setup

Data Sources

• Matroid Database: All matroids of small rank and few elements obtained from polyDB
• Computational Range:
  • Rank 2: All matroids up to 12 elements
  • Rank 3: All matroids up to 8 elements
  • Rank 4: 8 elements (from GHSV24) and 9 elements

Computational Tools

• Primary Software: Magma 2.27 for Gröbner basis computation
• Parallelization: GNU Parallel
• Hardware: 2×8-core Intel Xeon Gold 6144 @ 3.5GHz, 768GB RAM
• Open Source Implementation: Experimental code in the OSCAR system

Selection Criteria

1. Computationally check each matroid for half-coloops to filter self-projecting matroids
2. For each self-projecting matroid, select the isomorphic matroid with k as a basis
3. Further optimization: Select isomorphic matroids containing frames (for rank 3-8 elements and rank 4-9 elements)

Timeout Settings

• Standard timeout: 360 seconds
• Extended timeout: 7000 seconds (for some rank 4-9 element matroids)

Evaluation Metrics

• Realization Space Dimension: dim(R(M)) and dim(S(M))
• Inclusion Relations: S(M)=R(M), S(M)⊊R(M), S(M)=∅
• Completion Rate: Proportion of matroids for which computation terminates

Experimental Results

Main Results

1. Quantitative Statistics of Self-Projecting Matroids (Table 1)

Observations:

• In ranks 2 and 3, most matroids are self-projecting
• In rank 4, the proportion of self-projecting matroids decreases significantly
• The disjoint basis property (satisfied by 128,676 rank 4-9 element matroids) is more common than self-projectivity (7365 matroids)

2. Complete Classification for Rank 2 (Theorem 4.9)

Conclusion: All rank-2 self-projecting matroids M satisfy S(M)=R(M)

Proof Strategy:

• Rank-2 matroids are characterized by the set of circuits L and partition into parallel classes P_1,...,P_r
• Half-coloop exists ⟺ r∈{2,3} and |P_r|=1
• For r≥4, any realization can be transformed into a self-projecting realization by appropriate choice of λ
• Special cases with r=2,3 are verified individually

3. Rank-3 Results (Theorem 4.10, Table 2)

Key Findings:

• All rank-3 self-projecting matroids (≤8 elements) except U_{3,6} satisfy S(M)=R(M)
• U_{3,6} has S(M) as a codimension-1 subvariety of R(M)
• 4 matroids did not complete computation (dimension 4 or 5)
• By CL23, realization spaces of rank-3 matroids (≤11 elements) are smooth, so when S(M)=R(M), the self-projecting realization space is also smooth

4. Detailed Results for Rank 4-9 Elements (Theorem 4.11, Table 3)

Overall Statistics:

• Total self-projecting matroids: 7365 (including uniform matroids)
• Realizable matroids: 7181
• S(M)=R(M)≠∅: 174
• S(M)⊊R(M): At least 5400
• ∅≠S(M)⊊R(M): At least 2844
• S(M)=∅ (realizable but no self-projecting realization): At least 2556
• Computation did not terminate: 1606

Dimension Distribution:

Realization spaces R(M):

Self-projecting realization spaces S(M):

Realizable Matroids without Self-Projecting Realization (Table 4):

Ablation Study: Case Analysis

Case 4.12: Example without Self-Projecting Realization

The matroid is defined by linear dependencies of columns of the matrix: