OctoInformBeta

OpenAI preprint, September 2026 · Appendix A. A path-containment assertion

Appendix A (example): Failure of all-intermediate-times containment: On the open family of normalized real quadratics with negative leading coefficient, arbitrarily small deformations with fixed endpoints can shrink rigidly convex sets

Smooth symmetric determinantal quadratic paths can fail intermediate containment.

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

Given a normalized real quadratic polynomial with value 11 at the origin and negative leading coefficient, consider smooth one-parameter deformations that agree with the initial polynomial at t=0t = 0 and t=1t = 1 and admit a smooth monic real symmetric determinantal representation. The question is whether the associated interval around the origin at intermediate times t∈(0,1)t \in (0, 1) must contain the endpoint interval, or whether containment can fail at every t∈(0,1)t \in (0, 1) even for arbitrarily small deformations. These notes go through that example and the definitions it is built on.

Definitions

The rigidly convex set ℛ(p)⊂ℝ\mathcal{R}(p) \subset \mathbb{R} of a real polynomial p∈ℝ[x]p \in \mathbb{R}[x] is the subset of ℝ\mathbb{R} given by the closure of the connected component of 00 in {x∈ℝ:p(x)≠0}\{x \in \mathbb{R} : p(x) \neq 0\}.✦

Given nonzero real numbers aa and bb, the interval polynomial g∈ℝ[x]g \in \mathbb{R}[x] (written ga,bg_{a,b} when aa and bb vary) is the polynomial defined by g(x)=(1−xa)(1+xb).g(x) = \left(1 - \frac{x}{a}\right)\left(1 + \frac{x}{b}\right).(1)✦

For η∈ℝ\eta \in \mathbb{R}, the time scale c:ℝ→ℝc : \mathbb{R} \to \mathbb{R} is the real-valued function sending each t∈ℝt \in \mathbb{R} to c(t)=1+ηsin⁡2(πt).c(t) = 1 + \eta \sin^2(\pi t).(2)✦

The deformed polynomial Ht∈ℝ[x]H_t \in \mathbb{R}[x] (written Ht(η)H_t^{(\eta)} when η\eta varies), for nonzero real numbers a,ba, b and real numbers η,t\eta, t, is the real polynomial obtained from gg and cc by Ht(x)=g(c(t)x).H_t(x) = g(c(t)x).(3)✦

Given nonzero a,b∈ℝa, b \in \mathbb{R} and η,t∈ℝ\eta, t \in \mathbb{R}, the deformed coefficient vector h(t)∈ℝ3h(t) \in \mathbb{R}^3 of HtH_t is the triple of real numbers (h0(t),h1(t),h2(t))(h_0(t), h_1(t), h_2(t)) whose ii-th entry hi(t)h_i(t) for i∈{0,1,2}i \in \{0, 1, 2\} is the coefficient of xix^i in HtH_t.✦

For a natural number nn and a set RR, the set Mn(R)M_n(R) is the set of n×nn \times n matrices with entries in RR, the subspace SnS^n of Mn(ℝ)M_n(\mathbb{R}) is the subspace of real symmetric n×nn \times n matrices, and the identity matrix InI_n of Mn(ℝ)M_n(\mathbb{R}) is the n×nn \times n identity matrix.✦

In M2(ℝ)M_2(\mathbb{R}), for nonzero a,b∈ℝa, b \in \mathbb{R} and η,t,x∈ℝ\eta, t, x \in \mathbb{R}, the deformed pencil Mt(x)M_t(x) is the diagonal 2×22 \times 2 real square matrix defined by Mt(x)=diag⁡(1−c(t)xa,1+c(t)xb).M_t(x) = \operatorname{diag}\left(1 - \frac{c(t)x}{a},\; 1 + \frac{c(t)x}{b}\right).(4)✦

The Claim

Theorem 1. For any a,b>0a, b > 0, the polynomial gg satisfies {x∈ℝ:g(x)=0}={−b,a},\{x \in \mathbb{R} : g(x) = 0\} = \{-b, a\},(5)✦ with {x∈ℝ:g(x)=0} compact in ℝ,\{x \in \mathbb{R} : g(x) = 0\} \text{ compact in } \mathbb{R},(6)✦ and at every x∈ℝx \in \mathbb{R} with g(x)=0g(x) = 0, g′(x)≠0.g'(x) \neq 0.(7)✦ Under a,b>0a, b > 0, every x∈ℝx \in \mathbb{R} satisfies g′(−b)(x+b)g′(−b)b=1+xb\frac{g'(-b)(x + b)}{g'(-b)\,b} = 1 + \frac{x}{b}(8)✦ and g′(a)(x−a)−ag′(a)=1−xa,\frac{g'(a)(x - a)}{-a\,g'(a)} = 1 - \frac{x}{a},(9)✦ and the rigidly convex set satisfies ℛ(g)=[−b,a].\mathcal{R}(g) = [-b, a].(10)✦ For all nonzero a,b∈ℝa, b \in \mathbb{R} and all x∈ℝx \in \mathbb{R}, (1+xb)(1−xa)=g(x).\left(1 + \frac{x}{b}\right)\left(1 - \frac{x}{a}\right) = g(x).(11)✦

Let a,b∈ℝa, b \in \mathbb{R} be nonzero and let η∈ℝ\eta \in \mathbb{R}. For the deformed family, t↦h(t)∈C∞(ℝ,ℝ3),t \mapsto h(t) \in C^\infty(\mathbb{R}, \mathbb{R}^3),(12)✦ and for every t∈ℝt \in \mathbb{R}, Ht(0)=1.H_t(0) = 1.(13)✦ For the pencil MtM_t, (t,x)↦Mt(x)∈C∞(ℝ2,M2(ℝ)).(t, x) \mapsto M_t(x) \in C^\infty(\mathbb{R}^2, M_2(\mathbb{R})).(14)✦ Every t∈ℝt \in \mathbb{R} gives Mt(0)=I2,M_t(0) = I_2,(15)✦ and for all t,x∈ℝt, x \in \mathbb{R}, Mt(x)∈S2M_t(x) \in S^2(16)✦ and det⁡Mt(x)=Ht(x).\det M_t(x) = H_t(x).(17)✦ At t=0t = 0 and t=1t = 1, H0=gandH1=g,H_0 = g \quad \text{and} \quad H_1 = g,(18)✦ while at t=1t = 1, H1=1⋅g.H_1 = 1 \cdot g.(19)✦

Assume that a,b,η>0a, b, \eta > 0. For all t,x∈ℝt, x \in \mathbb{R} and z∈ℂz \in \mathbb{C}, Ht(xz)=0⟹Im⁡z=0.H_t(xz) = 0 \implies \operatorname{Im} z = 0.(20)✦ For any t∈ℝt \in \mathbb{R}, {x∈ℝ:Ht(x)=0}={−bc(t),ac(t)},\{x \in \mathbb{R} : H_t(x) = 0\} = \left\{-\frac{b}{c(t)},\; \frac{a}{c(t)}\right\},(21)✦ with −bc(t)<ac(t)-\frac{b}{c(t)} < \frac{a}{c(t)}(22)✦ and {x∈ℝ:Ht(x)=0} compact in ℝ.\{x \in \mathbb{R} : H_t(x) = 0\} \text{ compact in } \mathbb{R}.(23)✦ At each t∈ℝt \in \mathbb{R}, every x∈ℝx \in \mathbb{R} with Ht(x)=0H_t(x) = 0 satisfies Ht′(x)≠0,H_t'(x) \neq 0,(24)✦ and the rigidly convex set satisfies ℛ(Ht)=[−bc(t),ac(t)].\mathcal{R}(H_t) = \left[-\frac{b}{c(t)},\; \frac{a}{c(t)}\right].(25)✦ When t∈(0,1)t \in (0, 1), ℛ(Ht)⊊ℛ(g)\mathcal{R}(H_t) \subsetneq \mathcal{R}(g)(26)✦ and ℛ(g)⊈ℛ(Ht).\mathcal{R}(g) \not\subseteq \mathcal{R}(H_t).(27)✦

For any real polynomial p∈ℝ[x]p \in \mathbb{R}[x] with xix^i-coefficient pip_i (i≥0i \ge 0), deg⁡p=2,p0=1,p2<0⇔there exist a,b>0 with p=ga,b.\deg p = 2,\; p_0 = 1,\; p_2 < 0 \iff \text{there exist } a, b > 0 \text{ with } p = g_{a,b}.(28)✦ In the parameter space ℝ2\mathbb{R}^2, {(v1,v2)∈ℝ2:v2<0} is open in ℝ2,\{(v_1, v_2) \in \mathbb{R}^2 : v_2 < 0\} \text{ is open in } \mathbb{R}^2,(29)✦ and for every (v1,v2)∈ℝ2(v_1, v_2) \in \mathbb{R}^2 with v2<0v_2 < 0, there exist a,b>0a, b > 0 such that in ℝ[x]\mathbb{R}[x], 1+v1x+v2x2=ga,b(x).1 + v_1 x + v_2 x^2 = g_{a,b}(x).(30)✦ For every a,b,ε>0a, b, \varepsilon > 0 (setting g=ga,bg = g_{a,b}), there exists η∈(0,ε)\eta \in (0, \varepsilon) such that for all t∈(0,1)t \in (0, 1), ℛ(Ht(η))⊊ℛ(g).\mathcal{R}(H_t^{(\eta)}) \subsetneq \mathcal{R}(g).(31)✦

In the paper, the result reads: for a,b,η>0a, b, \eta > 0, gg has compact smooth zero set {−b,a}\{-b, a\}, normalized tangent linear factors 1+x/b1 + x/b and 1−x/a1 - x/a multiplying to gg, and rigidly convex set [−b,a][-b, a]; each HtH_t (0≤t≤10 \le t \le 1) depends smoothly on tt, satisfies Ht(0)=1H_t(0) = 1, is real-zero with distinct real roots −b/c(t)-b/c(t) and a/c(t)a/c(t), admits the smooth monic real symmetric determinantal representation Ht(x)=det⁡diag⁡(1−c(t)x/a,1+c(t)x/b)H_t(x) = \det \operatorname{diag}(1 - c(t)x/a, 1 + c(t)x/b), and equals gg at t=0,1t = 0, 1 (cofactor 11), yet its rigidly convex set [−b/c(t),a/c(t)][-b/c(t), a/c(t)] is strictly smaller than [−b,a][-b, a] for 0<t<10 < t < 1; and the parameters a,b>0a, b > 0 describe the open family of normalized real quadratics with negative leading coefficient, and η>0\eta > 0 may be arbitrarily small. Smoothness of the zero sets is expressed by nonvanishing derivatives in (7) and (24), the real-zero condition by (20), the dependence of HtH_t and MtM_t over all t∈ℝt \in \mathbb{R} rather than only 0≤t≤10 \le t \le 1 by (12) and (14), and the open family of normalized quadratics in ℝ2\mathbb{R}^2 coordinates by (29) and (30).

What Is Checked

The claim is stated over real polynomials p∈ℝ[x]p \in \mathbb{R}[x] with their rigidly convex sets ℛ(p)\mathcal{R}(p), the interval polynomial gg, the time scale cc, the deformed polynomial HtH_t with coefficient vector h(t)∈ℝ3h(t) \in \mathbb{R}^3, and the deformed 2×22 \times 2 real square matrix pencil Mt(x)∈M2(ℝ)M_t(x) \in M_2(\mathbb{R}) (along with real symmetric matrices S2S^2), for nonzero a,b∈ℝa, b \in \mathbb{R} and real η,t,x\eta, t, x as well as for positive a,b,η,ε>0a, b, \eta, \varepsilon > 0, intermediate times t∈(0,1)t \in (0, 1), complex numbers z∈ℂz \in \mathbb{C}, and coefficient pairs (v1,v2)∈ℝ2(v_1, v_2) \in \mathbb{R}^2 with v2<0v_2 < 0.