OctoInformBeta

OpenAI preprint, September 2026 · §6. The tangent obstruction (proof of Theorem 1.1)

§6: Tangent norm identity (6.1): Difference quotients on spherical tangent curves and a tangent-space relation for a unit vector with nonzero first coordinate

Normalized pencils satisfy a Choi-Lam tangent norm identity.

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 normalized pencil defined by positive unital linear maps on 4×44 \times 4 real symmetric matrices and a unit vector v∈ℝ4v \in \mathbb{R}^4. Evaluating kernel projections along unit-sphere curves through vv in orthogonal directions compares the Choi-Lam biquadratic form with Euclidean operator norms of difference quotients, posing the problem of obtaining a linear tangent norm identity at vv.

Normalized Pencils and Kernel Projections

For natural numbers mm and nn, let ℝn\mathbb{R}^n denote nn-dimensional real Euclidean space with inner product ⟨⋅,⋅⟩\langle \cdot, \cdot \rangle and Euclidean norm ∥⋅∥\|\cdot\|, and write ℝm×n\mathbb{R}^{m \times n} for the space of real m×nm \times n matrices (square matrices of order nn when m=nm = n). Denote by 𝒮n\mathcal{S}^n the subspace of real symmetric n×nn \times n matrices with identity matrix II (written InI_n when specifying the order), where A≽0A \succeq 0 means AA is positive semidefinite and A≻0A \succ 0 means AA is positive definite.✦ For any A∈ℝm×nA \in \mathbb{R}^{m \times n}, the Euclidean operator norm ∥A∥op\|A\|_{\mathrm{op}} is the operator norm of the linear map ℝn→ℝm\mathbb{R}^n \to \mathbb{R}^m defined by AA with respect to the Euclidean norms.✦

For a vector v∈ℝnv \in \mathbb{R}^n, the symmetric outer product vv⊤v v^\top of vv with itself is the matrix in 𝒮n\mathcal{S}^n whose entry in row ii and column jj (for 1≤i,j≤n1 \le i, j \le n) is vivjv_i v_j. For a square matrix A∈ℝn×nA \in \mathbb{R}^{n \times n}, the kernel projection proj⁡ker⁡A∈ℝn×n\operatorname{proj}_{\ker A} \in \mathbb{R}^{n \times n} is the matrix representing the orthogonal projection of ℝn\mathbb{R}^n onto ker⁡A\ker A.✦

For y=(y1,y2,y3)∈ℝ3y = (y_1, y_2, y_3) \in \mathbb{R}^3, the quadratic matrix Q(y)Q(y) is the matrix in ℝ3×3\mathbb{R}^{3 \times 3} whose entry in row ii and column jj (with indices taken cyclically in {1,2,3}\{1, 2, 3\}) is yi2+yi+12y_i^2 + y_{i+1}^2 when i=ji = j and −yiyj-y_i y_j when i≠ji \ne j.✦ For z,y∈ℝ3z, y \in \mathbb{R}^3, the Choi-Lam form b(z,y)b(z, y) is the biquadratic polynomial z⊤Q(y)zz^\top Q(y) z.✦ For y∈ℝ3y \in \mathbb{R}^3, the matrix map Φy:𝒮4→𝒮4\Phi_y : \mathcal{S}^4 \to \mathcal{S}^4 is the linear map sending each (1+3)×(1+3)(1 + 3) \times (1 + 3) block matrix (ar⊤rT)∈𝒮4\begin{pmatrix} a & r^\top \\ r & T \end{pmatrix} \in \mathcal{S}^4 (with a∈ℝa \in \mathbb{R}, r∈ℝ3r \in \mathbb{R}^3, and T∈𝒮3T \in \mathcal{S}^3) to (1): Φy(αr⊤rM)=(tr⁡(Q(y)M)−r⊤Q(y)−Q(y)rαQ(y)).\Phi_y \begin{pmatrix} \alpha & r^\top \\ r & M \end{pmatrix} = \begin{pmatrix} \operatorname{tr}(Q(y) M) & -r^\top Q(y) \\ -Q(y) r & \alpha Q(y) \end{pmatrix}.(1)✦

A normalized pencil (D,E,B)(D, E, B) with positive integer dimensions a,c≥1a, c \ge 1 consists of linear maps D:𝒮4→𝒮aD : \mathcal{S}^4 \to \mathcal{S}^a, E:𝒮4→𝒮cE : \mathcal{S}^4 \to \mathcal{S}^c, and B:ℝ3→ℝa×cB : \mathbb{R}^3 \to \mathbb{R}^{a \times c} such that DD and EE are positive (mapping positive semidefinite matrices to positive semidefinite matrices) and unital (D(I4)=IaD(I_4) = I_a and E(I4)=IcE(I_4) = I_c), and (D,E,B)(D, E, B) satisfies the Schur-complement equivalence (2): for all X,Z∈𝒮4X, Z \in \mathcal{S}^4 with X≻0X \succ 0 and all y∈ℝ3y \in \mathbb{R}^3, Z−Φy(X−1)≽0⇔E(Z)−B(y)⊤D(X)−1B(y)≽0.Z - \Phi_y(X^{-1}) \succeq 0 \iff E(Z) - B(y)^\top D(X)^{-1} B(y) \succeq 0.(2)✦

Given a normalized pencil (D,E,B)(D, E, B) with dimensions a,c≥1a, c \ge 1 and vectors v,u∈ℝ4v, u \in \mathbb{R}^4, the left kernel projection P(v)P(v) of (D,E,B)(D, E, B) at vv and the right kernel projection R(u)R(u) of (D,E,B)(D, E, B) at uu are the orthogonal projection matrices in ℝa×a\mathbb{R}^{a \times a} and ℝc×c\mathbb{R}^{c \times c} defined by (3): P(v)=proj⁡ker⁡D(I−vv⊤),R(u)=proj⁡ker⁡E(I−uu⊤).P(v) = \operatorname{proj}_{\ker D(I - v v^\top)}, \qquad R(u) = \operatorname{proj}_{\ker E(I - u u^\top)}.(3)

Tangent Curves and Difference Quotients

For vectors v=(v0,v′)v = (v_0, v') and u=(u0,u′)u = (u_0, u') in ℝ4\mathbb{R}^4 with v0,u0∈ℝv_0, u_0 \in \mathbb{R} and v′,u′∈ℝ3v', u' \in \mathbb{R}^3, the wedge coordinates v0u′−u0v′v_0 u' - u_0 v' of vv and uu is the vector in ℝ3\mathbb{R}^3 given by v0u′−u0v′v_0 u' - u_0 v'.✦ For any v∈ℝ4v \in \mathbb{R}^4, the tangent space TvT_v is the linear subspace of ℝ4\mathbb{R}^4 given by (4): Tv=v⟂={h∈ℝ4:⟨v,h⟩=0}.T_v = v^\perp = \{h \in \mathbb{R}^4 : \langle v, h \rangle = 0\}.(4)✦

The spherical tangent curve uv,h:ℝ→ℝ4u_{v,h} : \mathbb{R} \to \mathbb{R}^4 of vectors v,h∈ℝ4v, h \in \mathbb{R}^4 is defined at each parameter s∈ℝs \in \mathbb{R} by (5): uv,h(s)=v+sh1+s2∥h∥2.u_{v,h}(s) = \frac{v + s h}{\sqrt{1 + s^2 \|h\|^2}}.(5)✦

Let v∈ℝ4v \in \mathbb{R}^4 have norm ∥v∥=1\|v\| = 1 and take h∈Tvh \in T_v. Then for every s∈ℝs \in \mathbb{R}, the curve uv,hu_{v,h} satisfies (6): ∥uv,h(s)∥=1.\|u_{v,h}(s)\| = 1.(6)✦ For any vectors v,h∈ℝ4v, h \in \mathbb{R}^4, the curve uv,hu_{v,h} is differentiable at s=0s = 0, with derivative (7): dds|s=0uv,h(s)=h.\left.\frac{d}{ds}\right|_{s=0} u_{v,h}(s) = h.(7)✦ For any vectors v=(v0,v′)v = (v_0, v') and h=(h0,h′)h = (h_0, h') in ℝ4\mathbb{R}^4 and every s∈ℝs \in \mathbb{R}, setting u=uv,h(s)=(u0,u′)u = u_{v,h}(s) = (u_0, u'), the wedge coordinates satisfy (8): v0u′−u0v′=s1+s2∥h∥2(v0h′−h0v′).v_0 u' - u_0 v' = \frac{s}{\sqrt{1 + s^2 \|h\|^2}}(v_0 h' - h_0 v').(8)✦

Given a normalized pencil (D,E,B)(D, E, B) with dimensions a,c≥1a, c \ge 1, take v=(v0,v′)∈ℝ4v = (v_0, v') \in \mathbb{R}^4 with ∥v∥=1\|v\| = 1 and h=(h0,h′)∈Tvh = (h_0, h') \in T_v. Then for every y∈ℝ3y \in \mathbb{R}^3 and every s∈ℝs \in \mathbb{R} with s≠0s \ne 0, the difference quotient satisfies (9): b(v0h′−h0v′,y)1+s2∥h∥2=∥P(v)B(y)R(uv,h(s))−R(v)s∥op2.\frac{b(v_0 h' - h_0 v', y)}{1 + s^2 \|h\|^2} = \left\| P(v) B(y) \frac{R(u_{v,h}(s)) - R(v)}{s} \right\|_{\mathrm{op}}^2.(9)✦

The Claim

Theorem 1. Let (D,E,B)(D, E, B) be a normalized pencil with dimensions a,c≥1a, c \ge 1. Then there exist a unit vector v=(v0,v′)∈ℝ4v = (v_0, v') \in \mathbb{R}^4 with v0≠0v_0 \ne 0 and a linear map T:ℝ4→ℝc×cT : \mathbb{R}^4 \to \mathbb{R}^{c \times c} such that, for every h=(h0,h′)∈Tvh = (h_0, h') \in T_v and every y∈ℝ3y \in \mathbb{R}^3, b(v0h′−h0v′,y)=∥P(v)B(y)T(h)∥op2.b(v_0 h' - h_0 v', y) = \|P(v) B(y) T(h)\|_{\mathrm{op}}^2.(10)✦

In the paper, the result reads: for h=(h0,h′)∈v⟂h = (h_0, h') \in v^\perp, defining Jh=v0h′−h0v′Jh = v_0 h' - h_0 v' and us=(v+sh)/1+s2∥h∥2u_s = (v + s h)/\sqrt{1 + s^2 \|h\|^2} and passing to the limit s→0s \to 0 in the operator-norm identity yields b(Jh,y)=∥P(v)B(y)dRv(h)∥op2b(Jh, y) = \|P(v) B(y) dR_v(h)\|_{\mathrm{op}}^2 for all h∈v⟂h \in v^\perp and y∈ℝ3y \in \mathbb{R}^3, where TvT_v stands for v⟂v^\perp, uv,h(s)u_{v,h}(s) for usu_s, and dRvdR_v is the differential of RR at vv.

What Is Checked

The tangent norm identity (10) and the preceding claims are stated over Euclidean spaces ℝn\mathbb{R}^n, square matrices, real symmetric matrices 𝒮n\mathcal{S}^n, the outer product and symmetric outer product vv⊤v v^\top, wedge coordinates, the quadratic matrix Q(y)Q(y), the Choi-Lam form bb, the matrix map Φy\Phi_y, kernel projections, the Euclidean operator norm, the tangent space TvT_v, spherical tangent curves uv,hu_{v,h}, and a normalized pencil (D,E,B)(D, E, B) with positive dimensions a,c≥1a, c \ge 1 whose positive unital linear maps DD and EE and coupling map BB satisfy the Schur-complement equivalence, together with its left and right kernel projections P(v)P(v) and R(u)R(u). It also holds that for a unit vector v∈ℝ4v \in \mathbb{R}^4 and an orthogonal direction h∈Tvh \in T_v, the spherical tangent curve uv,h(s)u_{v,h}(s) has unit norm for all ss, has derivative hh at s=0s = 0, and satisfies the wedge coordinate identity (8), while the difference quotient of right kernel projections along the curve satisfies (9) for every nonzero parameter ss and every y∈ℝ3y \in \mathbb{R}^3.