OctoInform Beta

OpenAI preprint, September 2026 · §2. Rectangular tests from subspace expansion · §2

§2 (proof of Proposition 2.1): the shrunk-subspace obstruction (2.10): If ∑iAi⊗Zi\sum_i A_i\otimes Z_i is invertible for ZiZ_i of positive order, no vector in ker⁡∑iAi⊗Ti\ker\sum_i A_i\otimes T_i spans UU with dim⁡∑iTi(U)>ndim⁡U\dim\sum_i T_i(U)>n\dim U

An invertible pencil makes an expanding test injective.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

This item concerns the shrunk-subspace obstruction in the proof of Proposition 2.1 of the preprint: for a linear pencil of order qq over a field FF with coefficient matrices A0,…,An∈Matq(F)A_0,\dots,A_n\in\mathrm{Mat}_q(F), subspaces V,V−⊆FqV,V_-\subseteq F^q with dim⁡V−<dim⁡V\dim V_-<\dim V and AiV⊆V−A_iV\subseteq V_- for every ii, whose presence prevents the pencil from being invertible at matrices of positive order.

The Setting

Let FF be a field and A0,…,An∈Matq(F)A_0,\dots,A_n\in\mathrm{Mat}_q(F) the coefficient matrices of a linear pencil of order qq in homogeneous form. The test 𝒯\mathcal{T} of the pencil at linear maps T0,…,Tn:P→QT_0,\dots,T_n\colon P\to Q between vector spaces PP and QQ over FF is the linear map 𝒯=∑i=0nAi⊗Ti\mathcal{T}=\sum_{i=0}^{n}A_i\otimes T_i from PqP^q to QqQ^q that sends f=(f1,…,fq)f=(f_1,\dots,f_q) to the vector whose entry α\alpha is 𝒯(f)α=∑i=0n∑γ=1q(Ai)αγTi(fγ),α=1,…,q.\mathcal{T}(f)_\alpha=\sum_{i=0}^{n}\sum_{\gamma=1}^{q}(A_i)_{\alpha\gamma}\,T_i(f_\gamma), \qquad \alpha=1,\dots,q.(1)

The span of a vector f=(f1,…,fq)∈Pqf=(f_1,\dots,f_q)\in P^q is the subspace U=⟨f1,…,fq⟩U=\langle f_1,\dots,f_q\rangle of PP spanned by its entries, and the maps carry it to T0(U)+…+Tn(U)⊆QT_0(U)+\dots+T_n(U)\subseteq Q.✦

Block Evaluation and Coefficient Matrices

Fix Z0,…,Zn∈Mate(F)Z_0,\dots,Z_n\in\mathrm{Mat}_e(F). The Kronecker block matrix of A=(A0,…,An)A=(A_0,\dots,A_n) and Z=(Z0,…,Zn)Z=(Z_0,\dots,Z_n) is the matrix of order qeqe ℬ(A;Z)=∑i=0nAi⊗Zi.\mathcal{B}(A;Z)=\sum_{i=0}^{n}A_i\otimes Z_i .(2)✦ The block evaluation of AA at ZZ is the linear map ℰ(A;Z)\mathcal{E}(A;Z) on the space (Fq)e(F^q)^e of families v=(v1,…,ve)v=(v_1,\dots,v_e) of vectors of FqF^q that sends vv to the family with members ℰ(A;Z)(v)β=∑i=0n∑δ=1e(Zi)βδAivδ,β=1,…,e;\mathcal{E}(A;Z)(v)_\beta=\sum_{i=0}^{n}\sum_{\delta=1}^{e}(Z_i)_{\beta\delta}\,A_iv_\delta, \qquad \beta=1,\dots,e;(3)✦ up to the order of the basis, its matrix is ℬ(A;Z)\mathcal{B}(A;Z).

Given a basis b1,…,bkb_1,\dots,b_k of the span UU of f∈Pqf\in P^q, the coefficient matrix of ff in that basis is the matrix C∈Matq×k(F)C\in\mathrm{Mat}_{q\times k}(F) whose row α\alpha is the coordinate vector of fαf_\alpha, a vector of UU: fα=∑j=1kCαjbj,α=1,…,q.f_\alpha=\sum_{j=1}^{k}C_{\alpha j}\,b_j,\qquad \alpha=1,\dots,q.(4)✦ The concatenation of CC with the pencil A0,…,AnA_0,\dots,A_n is the q×(n+1)kq\times(n+1)k matrix [A0C∣A1C∣⋯∣AnC],\bigl[\,A_0C\mid A_1C\mid\cdots\mid A_nC\,\bigr],(5) whose column (i,j)(i,j) is column jj of AiCA_iC.

For vectors u1,…,uku_1,\dots,u_k of PP, the image combination map of T0,…,TnT_0,\dots,T_n at uu is the linear map F(n+1)k→QF^{(n+1)k}\to Q given by (cij)↦∑i=0n∑j=1kcijTi(uj).(c_{ij})\longmapsto\sum_{i=0}^{n}\sum_{j=1}^{k}c_{ij}\,T_i(u_j).(6)✦ At the basis b1,…,bkb_1,\dots,b_k of UU, its image is T0(U)+…+Tn(U)T_0(U)+\dots+T_n(U).

The Claim

Theorem 1. Let FF be a field, A0,…,An∈Matq(F)A_0,\dots,A_n\in\mathrm{Mat}_q(F), Z0,…,Zn∈Mate(F)Z_0,\dots,Z_n\in\mathrm{Mat}_e(F), and T0,…,Tn:P→QT_0,\dots,T_n\colon P\to Q linear maps between vector spaces over FF.

  1. Suppose that e≥1e\ge 1 and that the block evaluation ℰ(A;Z)\mathcal{E}(A;Z) is injective. If f∈Pqf\in P^q lies in the kernel of the test (1) and its span UU has dimension kk, then dim⁡(T0(U)+…+Tn(U))≤nk.\dim\bigl(T_0(U)+\dots+T_n(U)\bigr)\le nk.(7)✦

  2. Suppose that e≥1e\ge 1 and that ℰ(A;Z)\mathcal{E}(A;Z) is injective. If VV and V−V_- are subspaces of FqF^q with AiV⊆V−A_iV\subseteq V_- for i=0,…,ni=0,\dots,n, then dim⁡V≤dim⁡V−.\dim V\le\dim V_- .(8)✦

  3. Let f∈Pqf\in P^q lie in the kernel of the test, let b1,…,bkb_1,\dots,b_k be a basis of its span UU, and let CC be the coefficient matrix of ff in that basis. If dim⁡(T0(U)+…+Tn(U))>nk\dim\bigl(T_0(U)+\dots+T_n(U)\bigr)>nk, then rank⁡[A0C∣A1C∣⋯∣AnC]<k.\operatorname{rank}\bigl[\,A_0C\mid A_1C\mid\cdots\mid A_nC\,\bigr]<k.(9)✦

  4. If the block matrix ℬ(A;Z)\mathcal{B}(A;Z) is invertible, then ker⁡ℰ(A;Z)=0.\ker\mathcal{E}(A;Z)=0.(10)✦

In the paper, the result reads: a nonzero vector in the kernel of the test gives subspaces V,V−⊆FqV,V_-\subseteq F^q with dim⁡V=k\dim V=k, dim⁡V−<k\dim V_-<k and AiV⊆V−A_iV\subseteq V_- for every ii, which no substitution of matrices ZiZ_i with ∑iAi⊗Zi\sum_i A_i\otimes Z_i invertible allows.

What Is Checked

The claim is stated over the block matrix (2) and the block evaluation (3) of the coefficient matrices at Z0,…,ZnZ_0,\dots,Z_n, the test (1), the span UU of a vector f∈Pqf\in P^q with the coefficient matrix (4) of ff in a basis of UU and its concatenation (5), and the image combination map (6) at that basis.