\section{Real four-dimensional residue images} \label{sec:real-fourfold-images} We exclude the real, geometrically integral, full-span residue image of dimension four. The proof uses the quadratic equations of that image and the radial and skew identities of the actual transpose pencil. The conjugate-image alternative is treated separately. Fix a real regular anchor in the rank-eight case. Let $V=\R^{16}$, let $E\subset V$ be the eight-dimensional Jacobi image, and let $\iota:\R^8\to E$ and $K:\R^8\to E^\perp$ be orthogonal isomorphisms. Let $B$ be the real linear transpose pencil, and write \[ B:\R^8\longrightarrow\operatorname{Hom}(\R^8,V),\qquad G(a)=B(a)B(a)^T,\qquad L(a)=K^TB(a). \] The transpose-pencil construction supplies the following identities: \begin{equation}\label{eq:r4-pencil-inputs} \rank B(a)=8\quad(a\in\R^8\setminus\{0\}),\qquad B(a)e=-\iota(a),\qquad L(a)^T=-L(a), \end{equation} where $e\in\R^8$ is fixed. We also have a nondegenerate indefinite quadratic form $h(a)=a^THa$ on the coefficient space. In particular, $G(a)$ is positive semidefinite of rank eight at every nonzero real $a$. We use Lemma~\ref{lem:scalar-span-fixed-image} and Remark~\ref{rem:scalar-span-radial}: if the real span of the scalar quadratic entries of $G$ has dimension at most three, then \begin{equation}\label{eq:r4-alignment} \operatorname{im}B(a)=E\qquad(a\in\R^8\setminus\{0\}). \end{equation} This criterion is an input from the actual-pencil analysis, not an assertion about arbitrary symmetric quadratic matrices. All complex varieties below carry their reduced structure unless a scheme is specified explicitly. \begin{proposition}\label{prop:r4-real-image} Suppose that a geometrically integral real-defined fourfold $A\subset\mathbb P^7_{\mathbb C}$ spans $\mathbb P^7_{\mathbb C}$ and is contained in the common zero locus of $h$ and every entry of $G$. Then \eqref{eq:r4-alignment} holds. \end{proposition} We first record a consequence of skewness which will be used twice. \begin{lemma}\label{lem:r4-fixed-subspace} Let $U\subset\R^8$ have dimension at least four. If $\operatorname{im}B(a)$ is one fixed eight-space $T$ for all $a\in U\setminus\{0\}$, then $T=E$. \end{lemma} \begin{proof} The radial column in \eqref{eq:r4-pencil-inputs} gives $\iota(U)\subset T\cap E$. Therefore $R=K^TT$ has dimension $r\leq8-\dim U\leq4$. For every nonzero $a\in U$, the skew map $L(a)$ has image exactly $R$. A real skew map with image $R$ vanishes on $R^\perp$ and restricts to an invertible skew map on $R$. Thus $r\in\{0,2,4\}$. If $r=2$, the linear family $L|_U$ takes values in the one-dimensional space of skew maps supported on $R$. It has a nonzero kernel, contradicting its constant rank two. If $r=4$, then $\dim U=4$ and $a\mapsto L(a)|_R$ is injective. The Pfaffian on skew four-by-four matrices is \[ A_{12}A_{34}-A_{13}A_{24}+A_{14}A_{23}, \] a quadratic form of signature $(3,3)$. Its restriction to the four-dimensional image of $U$ would have no nonzero real zero, hence would be definite. A form of signature $(3,3)$ has no definite four-dimensional subspace. Therefore $r=0$, and $T=E$. \end{proof} \subsection{The quadratic base and its degree} Set \[ \mathcal V=\Span_{\R}\{h,\ G_{ij}:1\leq i,j\leq16\},\qquad Z(\mathcal V)=\{h=G_{ij}=0\}\subset\mathbb P^7_{\mathbb C}. \] This base has no real point, since $G(a)$ has rank eight at every nonzero real coefficient vector. \begin{lemma}\label{lem:r4-base-height} Under the hypotheses of Proposition~\ref{prop:r4-real-image}, either \eqref{eq:r4-alignment} holds or $Z(\mathcal V)$ has dimension four. In the latter case $\mathcal V$ contains a real regular sequence of three quadrics, and $\deg A\in\{4,6,8\}$. \end{lemma} \begin{proof} The inclusion of $A$ gives dimension at least four. A six-dimensional component would be the entire irreducible quadric $V(h)$, which has real points. Suppose that a component $Y$ has dimension five. Its span cannot have dimension at most five: this would make $Y$ a linear $\mathbb P^5$ contained in $V(h)$, whereas a nondegenerate complex eight-form has maximal isotropic vector dimension four. If $Y$ spans a hyperplane $\ell=0$, the restriction of $h$ to that hyperplane has rank at least six. Thus $Y=V(h,\ell)$, an irreducible quadric. For every quadratic entry $g$ of $G$, \[ g=c_g h+\ell m_g \] with $m_g$ linear. Restriction to $A$ gives $\ell m_g=0$. Its integral coordinate ring and full span imply first $m_g|_A=0$, then $m_g=0$. All entries of $G$ are multiples of $h$, impossible at a nonzero real $h$-null vector. It remains that $Y$ spans all $\mathbb P^7$. Its degree is at least three by the nondegenerate degree bound \cite[p.~1]{EGHP2004}. There is a real entry $f$ of $G$ independent of $h$, by the preceding real-null-vector argument. The complete intersection $V(h,f)$ has pure dimension five and degree four. If $Y\ne\overline Y$, its two conjugate components would have total degree at least six. Hence $Y$ is real-defined. A generic transverse real complementary plane meets it only in conjugate nonreal pairs, so its degree is even. Therefore $\deg Y=4$. The degree of the complete intersection is exhausted by $Y$ with generic multiplicity one. A complete intersection has no embedded associated primes, so generic reducedness makes it reduced. It is exactly $Y$ as a scheme. Its homogeneous ideal is saturated in this positive dimension; consequently every quadratic entry of $G$ lies in $\Span_{\R}\{h,f\}$. The small scalar-span criterion gives \eqref{eq:r4-alignment}. Otherwise the ideal generated by $\mathcal V$ has height three. Start with $h$, choose a real quadric independent of it, and choose a third real element outside the finitely many height-two associated primes of the first two. None contains all of $\mathcal V$; over the infinite real field, a finite union of proper linear subspaces cannot cover $\mathcal V$. This gives the required regular sequence. Its degree-eight complete intersection contains $A$. Since $A$ is nondegenerate of codimension three, $\deg A\geq4$. Its degree is even by the same conjugation argument. Thus its only possibilities are four, six, and eight. \end{proof} If $\deg A=8$, it exhausts the three-quadric complete intersection. The same unmixedness and generic-reducedness argument identifies the intersection with $A$ as a scheme. Hence all quadratic entries of $G$ lie in the span of its three defining quadrics, and \eqref{eq:r4-alignment} follows. We now treat the other two degrees. \subsection{The degree-four image} \begin{lemma}\label{lem:r4-degree-four} If $\deg A=4$, then \eqref{eq:r4-alignment} holds. \end{lemma} \begin{proof} The variety has minimal degree. By the Del Pezzo--Bertini classification, as stated in \cite[Theorem~0.1]{EGHP2004}, it is a rational normal scroll or a cone over the Veronese surface. A singular variety in this list has a nonempty intrinsic linear vertex. Since $A$ is real-defined, such a vertex would be real-defined and would contain a real point. This is impossible. Thus $A$ is smooth. The four positive scroll integers sum to four, so all are one, and $A$ is the Segre variety \[ \mathbb P^1\times\mathbb P^3 \subset\mathbb P(\mathbb C^2\otimes\mathbb C^4). \] Conjugation preserves its two rulings, which have different dimensions. It therefore lifts, up to a scalar, to $J\otimes K_0$ with conjugate-linear maps on the two factors. Absorb that scalar into one factor. Then $(J\otimes K_0)^2=I$ forces $J^2=\alpha I$ and $K_0^2=\beta I$ with $\alpha\beta=1$. Commuting each map with its square makes its scalar real. Rescaling normalizes both squares to signs. The ambient real structure has square one, so those signs agree. If both signs were positive, fixed vectors in the two factors would give a real Segre point. Therefore both signs are negative. The standard $\mathrm{SU}(2)$ action on the first factor commutes with its quaternionic real structure $J$. It induces a real action on the coefficient eight-space; its three quaternion units give that space a left $\mathbb H$-module structure. The quadratic ideal of the Segre embedding is \[ \bigwedge\nolimits^2(\mathbb C^2)^* \otimes\bigwedge\nolimits^2(\mathbb C^4)^*. \] Indeed it is the kernel of restriction of quadratic polynomials to rank-one tensors. The first $\mathrm{SL}(2,\mathbb C)$ acts trivially on this space. Every entry of $G$ is consequently $\mathrm{SU}(2)$-invariant. By homogeneity, \[ G(ba)=|b|^2G(a)\qquad(b\in\mathbb H). \] For every nonzero real $a$, the space $U=\mathbb H a$ has real dimension four, since nonzero quaternions act invertibly. The displayed identity makes $\operatorname{im}B$ one fixed eight-space on $U\setminus\{0\}$. Lemma~\ref{lem:r4-fixed-subspace} identifies it with $E$. Since $a$ was arbitrary, the conclusion holds on the entire coefficient space. This argument does not require the quaternion action to preserve the original coefficient metric. \end{proof} \subsection{The degree-six image} \begin{lemma}\label{lem:r4-residual-quadric} If $\deg A=6$ and alignment has not already been obtained, there is a reduced real complete intersection $C=A\cup Q$ of three quadrics, where $Q$ is a real-defined quadric fourfold spanning a real vector six-space $U$ on which $h$ is definite. \end{lemma} \begin{proof} The quadratic $t=\tr G$ is positive definite and belongs to $\mathcal V$. It is independent of the indefinite form $h$. The height-three conclusion of Lemma~\ref{lem:r4-base-height} allows a real $f\in\mathcal V$ such that $(h,t,f)$ is a regular sequence. Its complete intersection $C$ has degree eight and has no real points. Every four-dimensional irreducible component of $C$ has degree at least two: degree one would give a linear $\mathbb P^4$ in $V(h)$, contradicting the maximal isotropic dimension four. The component $A$ consumes degree six with multiplicity one. The remaining degree is exactly one component $Q$ of degree two and multiplicity one. Unmixedness and generic reducedness make $C=A\cup Q$ reduced. The unique residual component is preserved by conjugation. The degree bound puts the span of $Q$ in a projective five-space; it cannot be a projective four-space, since then $Q$ would be that linear space. Thus $Q$ is a quadric hypersurface in $\mathbb P(U_{\mathbb C})$ for a real six-space $U$. Its equation is proportional to $h|_U$, which is nonzero by the maximal isotropic dimension bound. Since $Q$ has no real point, this real quadratic form is definite. \end{proof} \begin{lemma}\label{lem:r4-four-quadrics} In Lemma~\ref{lem:r4-residual-quadric}, choose real independent linear forms $u,v$ with $U=\ker u\cap\ker v$. There are real quadrics $s,r,\Delta$ such that \[ I_{C}=(h,s,r),\qquad (I_A)_2\subset\Span_{\R}\{h,s,r,\Delta\},\qquad s|_U=r|_U=0. \] Here $(I_A)_2$ denotes the real degree-two part of the homogeneous ideal. \end{lemma} \begin{proof} Since $I_Q=(h,u,v)$, subtracting constant multiples of $h$ from $t,f$ gives \[ s=u a_0+v b_0,\qquad r=u c_0+v d_0,\qquad \Delta=a_0d_0-b_0c_0, \] where $a_0,b_0,c_0,d_0$ are real linear forms. Work in the graded domain $S=\R[x_1,\ldots,x_8]/(h)$ in coefficient coordinates $x_i$. Both $(s,r)$ and $(u,v)$ are regular sequences; the latter follows from $h|_U\ne0$ and the Cohen--Macaulay property of the polynomial ring. For $g\in(I_A)_2$, reducedness gives $ug,vg\in I_A\cap I_Q=I_C$. Modulo $h$, write \[ ug=a_1s+b_1r,\qquad vg=c_1s+d_1r \] with real linear coefficients. The Koszul syzygies of $(s,r)$, with degrees taken into account, give a real constant $k$ such that \[ va_1-uc_1=kr,\qquad vb_1-ud_1=-ks. \] Substitute the expressions for $s,r$ and use the Koszul syzygies of $(u,v)$. There are real constants $e_1,e_2$ with \[ a_1-kd_0=e_1u,\quad c_1+kc_0=e_1v,\qquad b_1+kb_0=e_2u,\quad d_1-ka_0=e_2v. \] It follows that \[ ug=k(d_0s-b_0r)+u(e_1s+e_2r) =u(k\Delta+e_1s+e_2r). \] Cancel $u$ in the domain $S$. Lifting back adds a constant multiple of $h$, proving the assertion. No general linkage theorem is needed. \end{proof} \begin{proof}[Completion of Proposition~\ref{prop:r4-real-image}] Only degree six remains. Lemma~\ref{lem:r4-four-quadrics} gives constant real symmetric matrices $D_1,D_2$ such that \[ G(b)=h(b)D_1+\Delta(b)D_2\qquad(b\in U). \] Choose the sign of $h|_U$ positive, absorbing it into $D_1$. On the connected ellipsoid $h(b)=1$, the continuous scalar function $\Delta$ has an interval as its image. Thus the midpoint of any two normalized Gram values is another such Gram value. All have rank eight and are positive semidefinite. The kernel of a sum of positive semidefinite matrices is the intersection of their kernels; the common rank therefore forces all these kernels to agree. Homogeneity makes $\operatorname{im}B(b)$ one fixed eight-space for every nonzero $b\in U$. Lemma~\ref{lem:r4-fixed-subspace} shows that this image is $E$, so $L|_U=0$. Linearity now gives $L(a)=u(a)L_1+v(a)L_2$ for constant real matrices. The polynomial \[ Q(a)=\tr\bigl(L(a)L(a)^T\bigr) \] is a homogeneous binary quadratic in $u,v$. On $A$, the identity $G=0$ gives $LL^T=K^TGK=0$, so $Q$ vanishes there. If $Q$ were nonzero, it would factor over $\mathbb C$ into two nonzero linear forms, possibly repeated. The integral coordinate ring of $A$ would force one factor to vanish on $A$, contrary to full complex span. Hence $Q$ is the zero polynomial. On real coefficients it is a sum of squares of real entries, so $L(a)=0$ for every $a$. The rank condition in \eqref{eq:r4-pencil-inputs} now gives \eqref{eq:r4-alignment}. \end{proof} \begin{remark}\label{rem:r4-generic-finish} For a residue image of type $(\alpha,s)=(4,4)$, the previously established rank identity $\rank G=2s-8$ gives exactly the vanishing hypothesis used above. Full real support becomes full complex span when this image is geometrically integral and real-defined. At anchors in this branch, the proposition supplies $K^TB=0$. The frame-free closed-kernel propagation from a nonempty real open branch, followed by Lemma~\ref{lem:m03-closed-kernel}, gives the Clifford conclusion. The residual quadric $Q$ in the degree-six proof is not assumed to be a residue image, and no transpose-rank condition on $Q$ is used. The two nonreal conjugate-image case requires its separate argument. \end{remark}