\section{Quadratic matrices with a small scalar span} \label{sec:scalar-span} \begin{lemma}\label{lem:scalar-span-fixed-image} Let $n\ge3$, and let $C(a)$ be a real symmetric homogeneous quadratic matrix on $\R^n$. Suppose that $C(a)$ is positive semidefinite of one fixed positive rank for every $a\ne0$. If its scalar quadratic entries span a real vector space of dimension at most three, then its image is independent of $a\ne0$. \end{lemma} \begin{proof} The quadratic form $q(a)=\tr C(a)$ is positive definite and belongs to the scalar-entry span. Complete it to a basis $q,p,r$, allowing unused forms to be zero. Thus $C=qD_0+pD_1+rD_2$ for constant symmetric matrices $D_i$. Change real coefficient coordinates so that $q(a)=\norm a^2$. Brickman's theorem states that the joint range of two real quadratic forms on a unit sphere in dimension at least three is convex; see \cite[Theorem~1.1]{MartinezLegaz2005}. Consequently $\{C(a):q(a)=1\}$ is convex. For any two matrices in this normalized family, their midpoint is another member of the same rank. For positive semidefinite matrices $A,B$ one has $\ker(A+B)=\ker A\cap\ker B$. Equal ranks therefore force the endpoint kernels to agree. This holds for every pair, and symmetry gives a common image. Homogeneity extends the conclusion from $q=1$ to all nonzero coefficients. \end{proof} \begin{remark}\label{rem:scalar-span-radial} For a full eight-coefficient transpose pencil $B(a)$ with $C(a)=B(a)B(a)^T$ of real rank eight and radial column $B(a)e=-La$, where $L$ maps the coefficient space isomorphically onto the Jacobi image $E$, the fixed image in Lemma~\ref{lem:scalar-span-fixed-image} is exactly $E$. Indeed, over the reals the images of $C(a)$ and $B(a)$ agree, and the radial vectors span $E$. The conclusion is an image statement; it does not by itself identify the entire Gram matrix with a radial scalar matrix. The closed-kernel criterion supplies that subsequent step in the actual curvature setting. \end{remark}