OctoInformBeta

OpenAI preprint, September 2026 · §5. A matrix path with prescribed endpoints · §5

Lemma 5.2 [Continuation modulo two]: Terminal zeros under compactness and nondegeneracy via cubical flag parity and row-equivariant paths on lower triangular matrices

Equivariant continuation modulo two for maps and matrix paths.

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.

The Setting

Take a natural number NN and a continuous path Φθ\Phi_\theta (0≤θ≤10 \le \theta \le 1) of maps from N×NN \times N real lower triangular matrices CC to N×NN \times N real symmetric matrices, starting from a prescribed quadratic map at θ=0\theta = 0. When Φθ\Phi_\theta is equivariant under row sign reversals, has a compact zero locus on [0,1][0, 1], and vanishes only at matrices with nonzero rows, must Φ1(C)=0\Phi_1(C) = 0 have a lower triangular solution CC?

Alongside this matrix continuation problem, for a natural number dd, the lemma formulates an equivariant continuation principle on ℝd+1\mathbb{R}^{d+1} under coordinate reflections that fix the (d+1)(d+1)-st coordinate, using orbit balls, free action loci, and affine orbit models. Instead of smooth manifold degree theory, it states a finite combinatorial piecewise-linear parity principle on cubical flags under coordinate reflections, together with terminal-zero continuation results on ℝd+1\mathbb{R}^{d+1} for equivariant maps with affine, locally approximate, or simple initial zero orbits.

Lower Triangular Matrices and Sign Actions

For a natural number NN, the lower triangular matrix space ℒN\mathcal{L}_N is the space of N×NN \times N real matrices CC, indexed on rows and columns by {1,…,N}\{1, \dots, N\}, whose entry CijC_{ij} vanishes whenever i<ji < j.✦ For a natural number NN, a matrix C∈ℒNC \in \mathcal{L}_N has nonzero rows if for every row index i∈{1,…,N}i \in \{1, \dots, N\} there exists a column index j∈{1,…,N}j \in \{1, \dots, N\} with Cij≠0C_{ij} \neq 0.✦

For a vector s=(s1,…,sN)∈ℝNs = (s_1, \dots, s_N) \in \mathbb{R}^N and a matrix C∈ℒNC \in \mathcal{L}_N, the row sign action Rs(C)∈ℒNR_s(C) \in \mathcal{L}_N is the lower triangular matrix defined entrywise by (1): (Rs(C))ij=siCij,1≤i,j≤N.(R_s(C))_{ij} = s_i C_{ij}, \qquad 1 \le i, j \le N.(1)✦

The sign conjugation Ψs(Z)∈ℝN×N\Psi_s(Z) \in \mathbb{R}^{N \times N} of an N×NN \times N real matrix ZZ by s=(s1,…,sN)∈ℝNs = (s_1, \dots, s_N) \in \mathbb{R}^N is the matrix specified by (2): (Ψs(Z))ij=siZijsj,1≤i,j≤N.(\Psi_s(Z))_{ij} = s_i Z_{ij} s_j, \qquad 1 \le i, j \le N.(2)✦

Cubical Flags and Orbit Models

For a natural number nn, the coordinate reflection group ℛn\mathcal{R}_n is the group of nn-tuples s=(s1,…,sn)∈{0,1}ns = (s_1, \dots, s_n) \in \{0, 1\}^n of reflection indicators, where 00 leaves a coordinate unreflected and 11 reflects it.✦

Given a group GG acting on a metric space XX, a point p0∈Xp_0 \in X, and R∈ℝR \in \mathbb{R}, the group orbit ball BG(p0,R)⊆XB_G(p_0, R) \subseteq X is the set {x∈X:∃g∈G,dist⁡(x,g⋅p0)<R}\{x \in X : \exists g \in G,\; \operatorname{dist}(x, g \cdot p_0) < R\}.✦ The free action locus XGfree⊆XX^{\mathrm{free}}_G \subseteq X of a group GG acting on a set XX is the set {x∈X:∀g∈G,g⋅x=x⟹g=1}\{x \in X : \forall g \in G,\; g \cdot x = x \implies g = 1\}.✦ Relative to p0∈Xp_0 \in X and R∈ℝR \in \mathbb{R} for an action of a group GG on a metric space XX, the orbit ball label gG,p0,R:X→Gg_{G,p_0,R} : X \to G is the map that sends each x∈BG(p0,R)x \in B_G(p_0, R) to an element g∈Gg \in G satisfying dist⁡(x,g⋅p0)<R\operatorname{dist}(x, g \cdot p_0) < R, and every x∉BG(p0,R)x \notin B_G(p_0, R) to 1∈G1 \in G.✦

In dimension d∈ℕd \in \mathbb{N}, the spatial inclusion map ι:ℝd→ℝd+1\iota : \mathbb{R}^d \to \mathbb{R}^{d+1} is the linear map appending 00 as the (d+1)(d+1)-st coordinate, ι(u)=(u,0)\iota(u) = (u, 0).✦ The spatial projection map π:ℝd+1→ℝd\pi : \mathbb{R}^{d+1} \to \mathbb{R}^d for d∈ℕd \in \mathbb{N} is the linear map dropping the (d+1)(d+1)-st coordinate, sending x=(x1,…,xd+1)∈ℝd+1x = (x_1, \dots, x_{d+1}) \in \mathbb{R}^{d+1} to (x1,…,xd)(x_1, \dots, x_d).✦ Given d∈ℕd \in \mathbb{N}, a subgroup H≤ℛd+1H \le \mathcal{R}_{d+1}, a real vector space VV with a linear representation ρ:H→GL(V)\rho : H \to \mathrm{GL}(V) and a linear isomorphism L:ℝd→∼VL : \mathbb{R}^d \xrightarrow{\sim} V, together with p0∈ℝd+1p_0 \in \mathbb{R}^{d+1}, R∈ℝR \in \mathbb{R}, and q0∈ℝdq_0 \in \mathbb{R}^d, the equivariant affine model AH,ρ,L,p0,R,q0:ℝd+1→VA_{H,\rho,L,p_0,R,q_0} : \mathbb{R}^{d+1} \to V is the map defined for x∈ℝd+1x \in \mathbb{R}^{d+1} by (3): AH,ρ,L,p0,R,q0(x)=ρ(s)(L(π(s−1⋅x)−q0)),s=gH,p0,R(x).A_{H,\rho,L,p_0,R,q_0}(x) = \rho(s)\bigl(L(\pi(s^{-1} \cdot x) - q_0)\bigr), \qquad s = g_{H,p_0,R}(x).(3)✦

A cubical flag FF in dimension n∈ℕn \in \mathbb{N} is a triple (b,c,σ)(b, c, \sigma) consisting of a base point b∈ℤnb \in \mathbb{Z}^n, a corner indicator c∈{0,1}nc \in \{0, 1\}^n, and a permutation σ∈Sn\sigma \in S_n of {1,…,n}\{1, \dots, n\}.✦ For a cubical flag F=(b,c,σ)F = (b, c, \sigma) in dimension n∈ℕn \in \mathbb{N} and k∈{0,…,n}k \in \{0, \dots, n\}, the integer flag vertex vk(F)∈ℤnv_k(F) \in \mathbb{Z}^n is the point whose coordinates 1≤i≤n1 \le i \le n are given by (4): (vk(F))i=2bi+{1,σ−1(i)≤k,2ci,σ−1(i)>k.(v_k(F))_i = 2 b_i + \begin{cases} 1, & \sigma^{-1}(i) \le k, \\ 2 c_i, & \sigma^{-1}(i) > k. \end{cases}(4) The real flag vertex pk(F)∈ℝnp_k(F) \in \mathbb{R}^n of a cubical flag FF in dimension n∈ℕn \in \mathbb{N} at k∈{0,…,n}k \in \{0, \dots, n\} is the point 12vk(F)\frac{1}{2} v_k(F). Given n∈ℕn \in \mathbb{N}, h∈ℝh \in \mathbb{R}, a cubical flag FF in dimension nn, and weights w=(w0,…,wn)∈ℝn+1w = (w_0, \dots, w_n) \in \mathbb{R}^{n+1}, the flag convex combination comb⁡h(F,w)∈ℝn\operatorname{comb}_h(F, w) \in \mathbb{R}^n is the vector defined by (5): comb⁡h(F,w)=h∑k=0nwkpk(F).\operatorname{comb}_h(F, w) = h \sum_{k=0}^n w_k p_k(F).(5)✦

For a cubical flag F=(b,c,σ)F = (b, c, \sigma) in dimension n∈ℕn \in \mathbb{N} and i∈{1,…,n}i \in \{1, \dots, n\}, the corner flip flip⁡i(F)\operatorname{flip}_i(F) is the cubical flag obtained from FF by replacing cic_i with 1−ci1 - c_i.✦ The adjacent cube crossing cross⁡i(F)\operatorname{cross}_i(F) of a cubical flag F=(b,c,σ)F = (b, c, \sigma) in dimension n∈ℕn \in \mathbb{N} along coordinate i∈{1,…,n}i \in \{1, \dots, n\} is the cubical flag obtained from FF by replacing cic_i with 1−ci1 - c_i and bib_i with bi+2ci−1b_i + 2 c_i - 1.✦ Given i,j∈{1,…,n}i, j \in \{1, \dots, n\} and a cubical flag F=(b,c,σ)F = (b, c, \sigma) in dimension n∈ℕn \in \mathbb{N}, the order swap swap⁡i,j(F)\operatorname{swap}_{i,j}(F) is the cubical flag (b,c,σ∘τij)(b, c, \sigma \circ \tau_{ij}), where τij∈Sn\tau_{ij} \in S_n transposes ii and jj.✦ When n≥1n \ge 1, the facet neighbor flag nbr⁡j(F)\operatorname{nbr}_j(F) of a cubical flag F=(b,c,σ)F = (b, c, \sigma) in dimension n∈ℕn \in \mathbb{N} across facet j∈{0,…,n}j \in \{0, \dots, n\} is the cubical flag defined by (6): nbr⁡j(F)={flip⁡σ(1)(F),j=0,cross⁡σ(n)(F),j=n,swap⁡j,j+1(F),1≤j≤n−1.\operatorname{nbr}_j(F) = \begin{cases} \operatorname{flip}_{\sigma(1)}(F), & j = 0, \\ \operatorname{cross}_{\sigma(n)}(F), & j = n, \\ \operatorname{swap}_{j, j+1}(F), & 1 \le j \le n - 1. \end{cases}(6)✦

The bottom seed flag Fseed(h,p)F_{\mathrm{seed}}(h, p) for d∈ℕd \in \mathbb{N}, a nonzero step h∈ℝh \in \mathbb{R}, and p=(p1,…,pd)∈ℝdp = (p_1, \dots, p_d) \in \mathbb{R}^d is the cubical flag in dimension d+1d + 1 with base (⌊p1/h⌋,…,⌊pd/h⌋,0)∈ℤd+1(\lfloor p_1/h \rfloor, \dots, \lfloor p_d/h \rfloor, 0) \in \mathbb{Z}^{d+1}, corner indicator (0,…,0)∈{0,1}d+1(0, \dots, 0) \in \{0, 1\}^{d+1}, and identity permutation id∈Sd+1\mathrm{id} \in S_{d+1}.✦ The mesh grid point mh(z)∈ℝnm_h(z) \in \mathbb{R}^n for n∈ℕn \in \mathbb{N}, h∈ℝh \in \mathbb{R}, and z∈ℤnz \in \mathbb{Z}^n is the point h2z\frac{h}{2} z. Given n∈ℕn \in \mathbb{N}, a subgroup H≤ℛnH \le \mathcal{R}_n, and an HH-invariant set DD of cubical flags in dimension nn, the invariant flag vertex set 𝒱H(D)⊆ℤn\mathcal{V}_H(D) \subseteq \mathbb{Z}^n is the HH-invariant subset {vk(F):F∈D,0≤k≤n}\{v_k(F) : F \in D,\; 0 \le k \le n\}.✦

The Claim

Theorem 1. In the first three parts, let d∈ℕd \in \mathbb{N}, let H≤ℛd+1H \le \mathcal{R}_{d+1} be a subgroup with sd+1=0s_{d+1} = 0 for all s∈Hs \in H, let VV be a finite-dimensional real normed vector space with an isometric linear representation ρ:H→GL(V)\rho : H \to \mathrm{GL}(V) and a linear isomorphism L:ℝd→∼VL : \mathbb{R}^d \xrightarrow{\sim} V, and let p∈ℝdp \in \mathbb{R}^d. Assume f:ℝd+1→Vf : \mathbb{R}^{d+1} \to V satisfies f(s⋅x)=ρ(s)(f(x))f(s \cdot x) = \rho(s)(f(x)) for all s∈Hs \in H and x∈ℝd+1x \in \mathbb{R}^{d+1}, every x∈ℝd+1x \in \mathbb{R}^{d+1} with xd+1=0x_{d+1} = 0 and f(x)=0f(x) = 0 lies in the orbit H⋅ι(p)H \cdot \iota(p), the slab zero locus {x∈ℝd+1:xd+1∈[0,1],f(x)=0}\{x \in \mathbb{R}^{d+1} : x_{d+1} \in [0, 1],\; f(x) = 0\} is compact, and every zero xx in this locus belongs to (ℝd+1)Hfree(\mathbb{R}^{d+1})^{\mathrm{free}}_H.

  1. If ff is continuous on ℝd+1\mathbb{R}^{d+1}, f(ι(p))=0f(\iota(p)) = 0, and the map u↦f(ι(u))u \mapsto f(\iota(u)) on ℝd\mathbb{R}^d has Fréchet derivative LL at pp, then ∃x∈ℝd+1,xd+1=1andf(x)=0.\exists x \in \mathbb{R}^{d+1}, \qquad x_{d+1} = 1 \quad\text{and}\quad f(x) = 0.(7)✦

  2. If ff is continuous on ℝd+1\mathbb{R}^{d+1}, a real radius RR with 0<R≤10 < R \le 1 satisfies dist⁡(g⋅ι(p),g′⋅ι(p))>8R\operatorname{dist}(g \cdot \iota(p), g' \cdot \iota(p)) > 8R for all distinct g,g′∈Hg, g' \in H, BH(ι(p),R)⊆(ℝd+1)HfreeB_H(\iota(p), R) \subseteq (\mathbb{R}^{d+1})^{\mathrm{free}}_H, and every x∈BH(ι(p),R)x \in B_H(\iota(p), R) with xd+1=0x_{d+1} = 0 satisfies ∥f(x)−A(x)∥≤12∥A(x)∥\|f(x) - A(x)\| \le \frac{1}{2} \|A(x)\| for A=AH,ρ,L,ι(p),R,pA = A_{H,\rho,L,\iota(p),R,p}, then ∃x∈ℝd+1,xd+1=1andf(x)=0.\exists x \in \mathbb{R}^{d+1}, \qquad x_{d+1} = 1 \quad\text{and}\quad f(x) = 0.(8)✦

  3. If ff is continuous on {x∈ℝd+1:xd+1∈[0,1]}\{x \in \mathbb{R}^{d+1} : x_{d+1} \in [0, 1]\}, a real radius R>0R > 0 satisfies dist⁡(g⋅ι(p),g′⋅ι(p))>8R\operatorname{dist}(g \cdot \iota(p), g' \cdot \iota(p)) > 8R for all distinct g,g′∈Hg, g' \in H, and f(x)=A(x)f(x) = A(x) for every x∈BH(ι(p),R)x \in B_H(\iota(p), R) with xd+1=0x_{d+1} = 0, where A=AH,ρ,L,ι(p),R,pA = A_{H,\rho,L,\iota(p),R,p}, then ∃x∈ℝd+1,xd+1=1andf(x)=0.\exists x \in \mathbb{R}^{d+1}, \qquad x_{d+1} = 1 \quad\text{and}\quad f(x) = 0.(9)✦

  4. Let d∈ℕd \in \mathbb{N}, let H≤ℛd+1H \le \mathcal{R}_{d+1} be a subgroup with sd+1=0s_{d+1} = 0 for all s∈Hs \in H, let DD be a finite HH-invariant set of cubical flags in dimension d+1d + 1, let VV be a finite-dimensional real normed vector space with a linear representation ρ:H→GL(V)\rho : H \to \mathrm{GL}(V) and a linear isomorphism L:ℝd→∼VL : \mathbb{R}^d \xrightarrow{\sim} V, and let p,q0∈ℝdp, q_0 \in \mathbb{R}^d and R,h>0R, h > 0 satisfy 2h<R2h < R and dist⁡(g⋅ι(p),g′⋅ι(p))>8R\operatorname{dist}(g \cdot \iota(p), g' \cdot \iota(p)) > 8R for all distinct g,g′∈Hg, g' \in H. Assume a:𝒱H(D)→Va : \mathcal{V}_H(D) \to V satisfies a(s⋅z)=ρ(s)(a(z))a(s \cdot z) = \rho(s)(a(z)) for all s∈Hs \in H and z∈𝒱H(D)z \in \mathcal{V}_H(D), the family {(1,a(vk(F)))}k≠j\{(1, a(v_k(F)))\}_{k \neq j} is linearly independent in ℝ×V\mathbb{R} \times V for every F∈DF \in D and j∈{0,…,d+1}j \in \{0, \dots, d+1\}, and {a(vk(F))}k≠i,j\{a(v_k(F))\}_{k \neq i, j} is linearly independent in VV for all F∈DF \in D and distinct i,j∈{0,…,d+1}i, j \in \{0, \dots, d+1\}. Assume further that mh(z)∈(ℝd+1)Hfreem_h(z) \in (\mathbb{R}^{d+1})^{\mathrm{free}}_H for all z∈𝒱H(D)z \in \mathcal{V}_H(D), that a(vk(F))=AH,ρ,L,ι(p),R,q0(mh(vk(F)))a(v_k(F)) = A_{H,\rho,L,\iota(p),R,q_0}(m_h(v_k(F))) whenever F∈DF \in D and k∈{0,…,d+1}k \in \{0, \dots, d+1\} satisfy (mh(vk(F)))d+1=0(m_h(v_k(F)))_{d+1} = 0 and mh(vk(F))∈BH(ι(p),R)m_h(v_k(F)) \in B_H(\iota(p), R), that Fseed(h,p)∈DF_{\mathrm{seed}}(h, p) \in D with nbr⁡d+1(Fseed(h,p))∉D\operatorname{nbr}_{d+1}(F_{\mathrm{seed}}(h, p)) \notin D, and that there exist weights α=(α0,…,αd+1)∈ℝd+2\alpha = (\alpha_0, \dots, \alpha_{d+1}) \in \mathbb{R}^{d+2} with αd+1=0\alpha_{d+1} = 0, αk>0\alpha_k > 0 for all k∈{0,…,d}k \in \{0, \dots, d\}, ∑k=0d+1αk=1\sum_{k=0}^{d+1} \alpha_k = 1, and comb⁡h(Fseed(h,p),α)=ι(q0)\operatorname{comb}_h(F_{\mathrm{seed}}(h, p), \alpha) = \iota(q_0). Then there exist F=(b,c,σ)∈DF = (b, c, \sigma) \in D, j∈{0,…,d+1}j \in \{0, \dots, d+1\} with nbr⁡j(F)∉D\operatorname{nbr}_j(F) \notin D, and weights w=(w0,…,wd+1)∈ℝd+2w = (w_0, \dots, w_{d+1}) \in \mathbb{R}^{d+2} with wk≥0w_k \ge 0 for all k∈{0,…,d+1}k \in \{0, \dots, d+1\}, ∑k=0d+1wk=1\sum_{k=0}^{d+1} w_k = 1, and wj=0w_j = 0 such that ∑k=0d+1wka(vk(F))=0and((j,σ−1(d+1),bd+1,cd+1)≠(d+1,d+1,0,0) or comb⁡h(F,w)∉BH(ι(p),R/2)).\sum_{k=0}^{d+1} w_k a(v_k(F)) = 0 \quad\text{and}\quad \Bigl(\bigl(j, \sigma^{-1}(d+1), b_{d+1}, c_{d+1}\bigr) \neq (d+1, d+1, 0, 0) \;\text{ or }\; \operatorname{comb}_h(F, w) \notin B_H(\iota(p), R/2)\Bigr).(10)✦

  5. Let N∈ℕN \in \mathbb{N}, ε>0\varepsilon > 0, and k=(k1,…,kN)∈ℝNk = (k_1, \dots, k_N) \in \mathbb{R}^N with ki>0k_i > 0 for all i∈{1,…,N}i \in \{1, \dots, N\}. Suppose Φ:ℝ×ℒN→ℝN×N\Phi : \mathbb{R} \times \mathcal{L}_N \to \mathbb{R}^{N \times N}, (θ,C)↦Φθ(C)(\theta, C) \mapsto \Phi_\theta(C), is continuous on [0,1]×ℒN[0, 1] \times \mathcal{L}_N, and for every θ∈[0,1]\theta \in [0, 1] and C∈ℒNC \in \mathcal{L}_N the matrix Φθ(C)\Phi_\theta(C) is real symmetric and satisfies Φθ(Rs(C))=Ψs(Φθ(C))\Phi_\theta(R_s(C)) = \Psi_s(\Phi_\theta(C)) for all s∈{−1,1}Ns \in \{-1, 1\}^N. Assume that Φ0(C)=2diag⁡(k1,…,kN)−ε−1CCT\Phi_0(C) = 2\operatorname{diag}(k_1, \dots, k_N) - \varepsilon^{-1} C C^T for all C∈ℒNC \in \mathcal{L}_N, that the zero locus {(C,θ)∈ℒN×[0,1]:Φθ(C)=0}\{(C, \theta) \in \mathcal{L}_N \times [0, 1] : \Phi_\theta(C) = 0\} is compact, and that every C∈ℒNC \in \mathcal{L}_N solving Φθ(C)=0\Phi_\theta(C) = 0 for some θ∈[0,1]\theta \in [0, 1] has nonzero rows. Then ∃C∈ℒN,Φ1(C)=0.\exists C \in \mathcal{L}_N, \qquad \Phi_1(C) = 0.(11)✦

In the paper, the result reads: for a smooth dd-dimensional manifold QQ without boundary and a smooth real vector bundle E→QE \to Q of rank dd, if sθs_\theta (0≤θ≤10 \le \theta \le 1) is a continuous homotopy of sections whose zero locus {(q,θ)∈Q×[0,1]:sθ(q)=0}\{(q, \theta) \in Q \times [0, 1] : s_\theta(q) = 0\} is compact, and if s0s_0 is smooth and has an odd number of zeros, all transverse, then s1s_1 has a zero.

What Is Checked

For d,N∈ℕd, N \in \mathbb{N}, claims (7), (8), (9), and (10) are stated over finite-dimensional real normed vector spaces with linear representations of subgroups of ℛd+1\mathcal{R}_{d+1} fixing the last coordinate, orbit balls, free action loci, equivariant affine models, and finite invariant cubical flag complexes with their vertices, mesh grid points, bottom seed flags, and facet neighbors, while claim (11) is stated over the lower triangular matrix space ℒN\mathcal{L}_N with continuous symmetric-matrix paths equivariant under row sign actions and sign conjugations together with the nonzero-rows condition.