\documentclass[11pt,a4paper]{article} \usepackage{amsmath,amsthm,amssymb,mathtools} \usepackage[margin=27mm]{geometry} \usepackage[hyperfootnotes=false,colorlinks=true,linkcolor=blue,citecolor=blue,urlcolor=blue]{hyperref} \setlength{\emergencystretch}{2em} \newcommand{\R}{\mathbb R} \newcommand{\Q}{\mathbb Q} \newcommand{\Z}{\mathbb Z} \newcommand{\PP}{\mathbb P} \newcommand{\RR}{\mathcal R} \newcommand{\PD}{\operatorname{PD}} \newcommand{\diag}{\operatorname{diag}} \newcommand{\pind}{p_+} \newtheorem{theorem}{Theorem} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{proposition}[theorem]{Proposition} \usepackage{indentfirst} \title{A Counterexample to the Realizability Conjecture for Surface Classes} \author{Qiyue Tang} \date{15 November 2026} \begin{document} \maketitle \begin{abstract} We give a counterexample to Huang--Huh--Micha\l ek--Wang--Wang's conjecture on real realizability of universally pseudoeffective surface classes. \end{abstract} \begingroup \renewcommand{\thefootnote}{\relax} \footnotetext{\textbf{Keywords.} Surface classes, universal pseudoeffectivity, real realizability.\\ \textbf{2020 Mathematics Subject Classification.} Primary 14C25. Secondary 14C17.} \endgroup \section{Introduction} The algebraic Steenrod problem asks which homology classes are represented by irreducible algebraic subvarieties, possibly after multiplication by a positive integer. For a product of two projective spaces, Huh \cite[Theorem 21]{Huh} related this question to log-concavity of the multidegree. The multidegree records the degrees obtained by intersecting the subvariety with complementary numbers of hyperplanes from the two factors. Apart from the coordinate classes singled out in his theorem, the criterion is that the sequence of degrees be nonzero and nonnegative, with no zero between two positive entries, and that the square of each intermediate degree be at least the product of its neighbors. For surface classes, the Hodge index theorem gives a restriction on the intersection form induced by any irreducible surface. Babaee and Huh \cite[Proposition 5.9]{BabaeeHuh} used the persistence of this restriction under limits in their study of approximation by integration currents on smooth projective toric varieties. On the positivity side, Fulger and Lehmann \cite[\S 2.3]{FulgerLehmann} introduced universal pseudoeffectivity for numerical cycle classes. This condition requires pseudoeffectivity after pullback along arbitrary projective test morphisms. Huang, Huh, Micha\l ek, Botong Wang and Shouda Wang use its homological counterpart in their realization problem. Huang et al.\ \cite[Theorem 1.8]{HHMWW} characterize rationally realizable surface classes on products of projective spaces. The entries of their matrix are the intersection degrees with pairs of hyperplane classes pulled back from the factors. Their criterion is that the matrix have nonnegative entries and at most one positive eigenvalue. The corresponding real statement follows by rational approximation as in \cite[\S 7]{HHMWW}. They propose a corresponding criterion on arbitrary smooth complex projective varieties in Conjecture 7.3 of the May 2025 preprint. Let $X$ be a smooth complex projective variety of dimension $n\ge 2$. We use Betti homology and cohomology. For a reduced irreducible subvariety $S\subset X$ of dimension $k$, write $[S]$ for its fundamental class pushed to $H_{2k}(X,\R)$. For $0\le k\le n$, let $B_k(X)$ be the real span of these classes. An effective class is a finite sum of such classes with nonnegative coefficients. Integral and rational effective classes have coefficients in $\Z$ and $\Q$, respectively. The pseudoeffective cone is the closure of the cone of effective classes in real homology. Let $\RR_{2,\Q}(X)$ be the set of classes $a[S]$ with $a\in\Q_{>0}$ and $S\subset X$ a reduced irreducible surface. Its closure $\RR_2(X)$ in $B_2(X)$ consists of the classes realizable over $\R$ in the sense of \cite[Definition 1.5 and \S 7]{HHMWW}. In particular, $[S]/m\to0$ as the positive integer $m$ tends to infinity. For a smooth projective variety $U$, let $\PD_U$ denote the inverse of cap product with its fundamental class $[U]$. A lower star on a map denotes homological pushforward. For a morphism $\pi:Z\to X$ from a smooth projective variety of dimension $m$, the homological pullback of $\xi\in B_k(X)$ is \[ \pi^*\xi=\pi^*(\PD_X\xi)\frown[Z]\in H_{2(m-n+k)}(Z,\R). \] It is zero in negative homological degrees. A class is universally pseudoeffective if this pullback is pseudoeffective for every such morphism. Pullback and duality are taken componentwise for a disconnected source. Write $H^{1,1}(X)_\R$ for the real classes in the $(1,1)$ Hodge summand of $H^2(X,\mathbb C)$. For a surface class $\xi$ and $\alpha,\beta\in H^{1,1}(X)_\R$, set \[ q_\xi(\alpha,\beta)=\langle\alpha\smile\beta,\xi\rangle, \] where $\smile$ is cup product and the brackets denote evaluation on homology. Write $\pind(q)$ for the positive index of a real symmetric form $q$, the maximal dimension of a positive-definite subspace. By Sylvester's law of inertia it counts the strictly positive eigenvalues of a matrix of the form. Conjecture 7.3 in \cite{HHMWW} asserts that, for universally pseudoeffective $\xi\in B_2(X)$, \[ \xi\in\RR_2(X)\quad\Longleftrightarrow\quad\pind(q_\xi)\le1. \] \begin{theorem}\label{thm:main} There is a smooth complex projective fourfold $X$ and an integral effective class $\eta\in B_2(X)$ which is universally pseudoeffective and for which $q_\eta$ is nondegenerate with one positive and three negative eigenvalues, but $\eta\notin\RR_2(X)$. \end{theorem} The real statement for Grassmannians follows from the integral classification in \cite[Theorem 7.7]{HHMWW}. Their Example 7.4 shows that ordinary pseudoeffectivity is insufficient even with the Hodge index bound. Example 7.5 gives a universally pseudoeffective class satisfying that bound which is realizable over $\R$ but fails rational realizability. Example 7.6 asks about real realizability for the class of the Fano matroid, which satisfies both hypotheses and fails rational realizability. The obstruction comes from a finite flat double cover of $X$. The inverse image of an irreducible surface has at most two irreducible components, which bounds the positive index of the pulled-back form even after passage to the real closure. \section{The positive-index bound for finite flat covers} \begin{lemma}\label{lem:linear} For symmetric forms $q_1,\ldots,q_r$ on a fixed finite-dimensional real vector space, \[ \pind\left(\sum_{j=1}^r q_j\right)\le\sum_{j=1}^r\pind(q_j). \] For a fixed integer $d\ge0$, the set of symmetric forms with positive index at most $d$ is closed. \end{lemma} \begin{proof} Choose a subspace $N_j$ of codimension $\pind(q_j)$ on which $q_j$ is nonpositive, using a diagonalization of $q_j$. The intersection $N=\bigcap_jN_j$ has codimension at most the sum of these codimensions, and the sum form is nonpositive on $N$. A positive-definite subspace intersects $N$ only in zero, proving the inequality. For closedness, if a limit form is positive definite on a $(d+1)$-dimensional subspace, its restriction has a positive minimum on the unit sphere of that subspace, so all sufficiently close approximants are positive definite there as well. This contradicts their positive-index bound. \end{proof} \begin{lemma}\label{lem:surface} If $T$ is an irreducible surface in a smooth complex projective variety $Y$, then $\pind(q_{[T]})\le1$ on the entire space $H^{1,1}(Y)_{\R}$. \end{lemma} \begin{proof} Take a projective resolution $r:\widetilde T\to T\hookrightarrow Y$. It has $r_*[\widetilde T]=[T]$, so the projection formula gives \[ q_{[T]}(\alpha,\beta)=\int_{\widetilde T}r^*\alpha\smile r^*\beta. \] On $H^{1,1}(\widetilde T)_{\R}$ an ample class has positive square, and its orthogonal complement is negative definite by the Hodge--Riemann bilinear relations. The map $r^*$ is injective on any subspace on which the pulled-back form is positive definite, and its image there is positive definite for the surface pairing. \end{proof} Lemmas \ref{lem:linear} and \ref{lem:surface} show that every real-realizable class satisfies the conjectured positive-index bound on $X$ itself. \begin{proposition}\label{prop:cover} Let $f:Y\to X$ be a finite flat surjection of degree $d$ between smooth complex projective varieties. Then \[ \xi\in\RR_2(X)\quad\Longrightarrow\quad \pind\bigl(q_{f^*\xi}\text{ on }H^{1,1}(Y)_{\R}\bigr)\le d. \] \end{proposition} \begin{proof} First let $S\subset X$ be an arbitrary reduced irreducible surface. Its scheme-theoretic inverse image $T=Y\times_XS$ is finite flat of rank $d$ over $S$, and may be nonreduced. Each irreducible component of $T$ dominates $S$: flat going-down says that the contraction of a minimal prime is a minimal prime, and $S$ is integral. The algebra of the generic fiber has dimension $d$ over the function field of $S$ and is the product of its local Artinian factors. For each local factor, indexed by the generic point of an irreducible component $j$,, let $m_j$ be its module length and $e_j$ the degree of its residue field over the function field of $S$. Each composition factor of that local algebra is its residue field, so its vector-space dimension is $m_je_j$. Thus $m_j,e_j\ge1$ and $\sum_jm_je_j=d$. Consequently $T$ has $r\le d$ reduced irreducible components $T_j$, all of dimension two. Flat pullback of cycles, compatible with the homological pullback defined above, gives \[ f^*[S]=\sum_{j=1}^r m_j[T_j]. \] The multiplicities are positive; they equal one in the \'etale case. Lemmas \ref{lem:linear} and \ref{lem:surface} imply $\pind(q_{f^*[S]})\le r\le d$. Positive scalar multiplication leaves this bound unchanged. See \cite[Tag 02RA]{Stacks} for the flat cycle pullback used here. Finally, in any fixed basis of $H^{1,1}(Y)_{\R}$, every matrix entry of $q_{f^*\xi}$ is linear and continuous in $\xi\in B_2(X)$. Apply the closedness in Lemma \ref{lem:linear} to a sequence from $\RR_{2,\Q}(X)$ converging to $\xi$. \end{proof} \begin{lemma}\label{lem:gg} On a smooth projective variety, a product of first Chern classes of globally generated line bundles is Poincar'e dual to the homology class of an effective cycle, possibly zero. The same assertion holds after arbitrary pullback to a smooth projective variety. \end{lemma} \begin{proof} Choose sections successively. At each stage, global generation allows a section that vanishes identically on none of the finitely many components of the preceding intersection. The resulting divisors intersect properly, or the intersection is empty. Their intersection cycle is effective and represents the product of their first Chern classes. If there are more factors than the dimension, the class is zero. Pullback preserves global generation, so the same argument applies to each pullback. Disconnected test varieties can be treated componentwise. \end{proof} \section{The fourfold counterexample} \subsection{Descent of universal pseudoeffectivity} \begin{lemma}\label{lem:flatdescent} Let $f:Y\to X$ be finite flat and surjective of degree $d$, with $X,Y$ smooth projective, and let $\xi\in B_2(X)$. Suppose that every pullback of $f^*\xi$ to a smooth projective test variety is a rational effective homology class. Then the same is true of $\xi$. If all those upstairs pullbacks are integral effective, then $d\pi^*\xi$ is integral effective for every test morphism $\pi:Z\to X$. \end{lemma} \begin{proof} Fix a test morphism $\pi:Z\to X$ and work on one connected component of the smooth source at a time. The fiber product $W=Z\times_XY$ is finite flat of rank $d$ over $Z$, but need not be smooth or reduced. Resolve its reduced irreducible components projectively. Denote the resulting maps by \[ g_j:\widetilde W_j\to Z,\qquad h_j:\widetilde W_j\to Y, \qquad fh_j=\pi g_j. \] Every $g_j$ is generically finite and surjective. Let $m_j$ be the length of the local ring of $W$ at the generic point of the $j$th component and $e_j$ the degree of $g_j$. Resolution preserves the function field, so the generic-fiber dimension count in Proposition \ref{prop:cover} gives $\sum_jm_je_j=d$. Functoriality of cohomological pullback, interpreted in homology by Poincar\'e duality, and the projection formula between smooth varieties of equal dimension give \[ h_j^*f^*\xi=g_j^*\pi^*\xi,\qquad (g_j)_*g_j^*\pi^*\xi=e_j\pi^*\xi. \] Indeed $(g_j)_*[\widetilde W_j]=e_j[Z]$, which proves the second formula by taking the cap product with $\PD_Z(\pi^*\xi)$. Consequently \[ d\pi^*\xi=\sum_jm_j(g_j)_*h_j^*f^*\xi. \] Every summand is rational effective, or integral effective under the stronger hypothesis. Proper pushforward preserves effectivity, and division by $d$ proves the assertion. \end{proof} \begin{proof}[Proof of Theorem \ref{thm:main}] Let $V$ be the blowup of $\PP^2$ at $p_1=[1:0:0]$ and $p_2=[0:1:0]$. Denote by $H$ the pullback of the hyperplane class and by $E_1,E_2$ the classes of the exceptional curves. We write a divisor and its first Chern class with the same symbol when taking cohomological products. On $V$, a dot between divisor classes denotes the degree of their product. The intersection matrix of $H,E_1,E_2$ is $\diag(1,-1,-1)$: a general line misses both blown-up points, the exceptional curves are disjoint, and each has normal bundle $\mathcal O_{\PP^1}(-1)$. For indices in $\{1,2,3\}$, put \[ C_1=H,\qquad C_2=H-E_1,\qquad C_3=H-E_2. \] The pullback of the line system gives global generation of $C_1$. The line pencils through $p_1$ and $p_2$ give global generation of $C_2$ and $C_3$ after blowing up their respective base points. The intersection matrix gives $C_i\cdot C_j=1$ for $i\ne j$. Choose curves $L_1,L_2,L_3$ of these classes by taking strict transforms of the lines in homogeneous coordinates $[x:y:z]$ given by $x+y+z=0$, $y+z=0$, and $x+2z=0$, respectively. The pairwise intersections in the projective plane are $[0:-1:1]$, $[-2:1:1]$, and $[-2:-1:1]$. They avoid the centers, so the strict transforms meet transversely at the corresponding points. Let $\operatorname{pr}_1,\operatorname{pr}_2:V\times V\to V$ be the projections and let $\Delta\subset V\times V$ be the diagonal. Blow up $\Delta$ to obtain $b:Y\to V\times V$ with exceptional divisor $D$. The factor swap lifts to an involution $\sigma$ of $Y$. Its projective quotient $f:Y\to X$ is the Hilbert square of $V$, the scheme of length-two subschemes of $V$ \cite[\S 2.1]{HilbertSquare}. The quotient is smooth and $f$ is finite flat of degree two. Indeed, in common local analytic coordinates $x,y\in\mathbb C^2$ on the two factors near the diagonal, use $u=(x+y)/2$ and $v=(x-y)/2$. In one blowup chart put $v_1=r$ and $v_2=rw$. The other chart interchanges $v_1$ and $v_2$. The action fixes $u,w$ and sends $r$ to $-r$. The quotient has coordinates $(u,r^2,w)$, and the local ring upstairs is free over the invariant ring with basis $1,r$. This description also holds for the completed local rings, so flatness follows at the branch points by faithful flatness of completion. Away from $D$ the involution acts freely and the quotient map is \'etale of degree two. For $i\ne j$ let $T_{ij}$ be the strict transform of $L_i\times L_j$, and for $i