Intrinsic signs and lower bounds in real algebraic geometry (Q2450177)

A classical result in complex algebraic geometry says that any smooth cubic nonsingular surface in \(\mathbb{P}^3(\mathbb{C})\) contains 27 lines. The perhaps most elementary proof of this (but still no easy reading) can be found in [\textit{M. Reid}, Undergraduate algebraic geometry. Cambridge (UK) etc.: Cambridge University Press (1988; Zbl 0701.14001)] (whose proof goes through as far as the reviewer sees, if the surface is irreducible). A \textit{real} cubic surface in \(\mathbb{P}^3(\mathbb{R})\) contains either 3, 7, 15 or 27 real lines, which according to \S 27 of \textit{B. Segre} [The non-singular cubic surfaces: a new method of investigation with special reference to questions of reality. Oxford, Clarendon Press. London, Oxford University Press (1942; Zbl 0061.36701; JFM 68.0358.01)] are either elliptic or hyperbolic. (If the intersection of the surface with a plane contains a line the remaining intersection is typically a conic which defines on the line an elliptic or hyperbolic involution in the sense of projective geometry; see \textit{H. S. M. Coxeter} [Non-Euclidean geometry. 6th ed. Washington, DC: The Mathematical Association of America (1998; Zbl 0909.51003)]. More ideas on Segre's results can be gleaned from the introduction to the paper by \textit{R. Benedetti} and \textit{R. Silhol} [Topology 34, No. 3, 651--678 (1995; Zbl 0996.57519)] who noted that the types of lines found by Segre are also differentiated on the the homology level. The rest of the review follows closely the abstract and introduction to the present paper: The authors are possibly among the first to note that in all real cases the number of hyperbolic lines exceeds the number of elliptic ones always by 3. Inspired by Segre's classification, the authors (a) explain a general principle which leads to lower bounds in real algebraic geometry; (b) explain the reason for the appearance of intrinsic signs in the problem treated by Segre and show that the same phenomenon occurs in a large class of enumerative problems in real algebraic geometry; finally they (c) illustrate the two principles in the enumerative problem of counting real lines in real hypersurfaces of degree \(2n-5\) in \(\mathbb{P}^{n-1}(\mathbb{R})\); the classical case being \(n=4\). To this aim, Section 1 introduces the canonical Euler class of a real vector bundle; see [\textit{E. H. Spanier}, Algebraic topology. New York etc.: McGraw-Hill Book Company (1966; Zbl 0145.43303)]. The definition does not need orientation data (`intrinsic definition'). Relatively oriented bundles on smooth manifolds and the concept of degree for relatively oriented bundles over closed manifolds are introduced; this is a much weaker condition than what is usually required. In Section 2 a lower estimate for the number of zeros of a section in a relatively oriented bundle over a closed manifold is given as \(\sum_{x\in Z(s)} \mathrm{mult}_x(s) \geq |\mathrm{deg}|(\xi)\). Here a classical inequality of singularity theory due to \textit{D. Eisenbud} and \textit{H. I. Levine} [Ann. Math. (2) 106, 19--44 (1977; Zbl 0398.57020)] is used and the new result can be viewed as a global version of the classical one. Section 3 applies the general theory to the problem of counting real lines in a hypersurface \(Z(f)\subset P(V)\) defined by a homogeneous polynomial of degree \(2n-5.\) These lines correspond bijectively to the zeros of the section \(s_f \in \Gamma(S^{2n-5}(U^\vee))\) associated with \(f\). Here, \(U\) is the tautological plane bundle of the Grassmannian \(G_2(V)\) and the section \(s_f\) is obtained by restricting \(f\) to the linear 2-planes \(W\subset V\). A fundamental remark is that the bundle \(S^{2n-5}(U^\vee)\) is canonically relatively oriented so that to any regular zero of \(s_f\) one can assign a sign in an intrinsic way. Note that both the Grassmannian \(G_2(V)\) and the bundle \(S^{2n-5}(U^\vee)\) are non-orientable when \(n\) is odd. For the canonical Euler class of \(S^{2n-5}(U^\vee)\) one obtains \(e_{S^{ 2n-5}(U^\vee)} =1\cdot 3 \cdot \cdots \cdot (2n-5)=: (2n-5)!!\). This result depends on certain orientation conventions in linear algebra: one has to fix the standard orientations of the symmetric power \(S^k(T)\) and of the morphism space \(\mathrm{Hom}(T',T'')\) given oriented vector spaces \(T,T',T''\). Thus as a corollary the lower estimate \(\sum_{l\in \mathcal{R}} \mathrm{mult}_l(s_f) \geq (2n-5)!!\) for the number of lines on the real hypersurface \(Z(f)\) is obtained. Section 4 deals with the intrinsic sign of a regular real line \(l\) on a real hypersurface \(Z(f)\) w.r.t. the canonical relative orientation of \(S^{2n-5}(U^\vee)\). For this the intrinsic derivative of the section \(s_f\) at a point in the zero locus is computed explicitly and a general formula for its determinant is given. In the case \(n=4\) considered by Segre, an alternative geometric interpretation is obtained showing that the intrinsic signs assigned to regular real lines using the general formalism and the canonical relative orientation of the bundle \(S^{2n-5}(U^\vee)\) corresponds to the two kinds of lines (hyperbolic and elliptic) defined by Segre. The computations here boil down to the computation of the determinant of a certain Sylvester matrix, and hence to resultant calculations (of two binary quadratic forms). The article's introduction closes with presenting remarks on the contemporary context of the results. Here results of \textit{A. Degtyarev} and \textit{V. Kharlamov} [Russ. Math. Surv. 55, No. 4, 735--814 (2000; Zbl 1014.14030); translation from Usp. Mat. Nauk 55, No. 4, 129--212 (2000)] are cited according to which the number of real rational plane cubics through 8 points in \(\mathbb{P}^2(\mathbb{R})\) is at least 8. (The referee ventures to ask: is there a relation to the classical Cayley Bacharach theorem, formulation CB3, in [\textit{D. Eisenbud} et al., Bull. Am. Math. Soc., New Ser. 33, No. 3, 295--324 (1996; Zbl 0871.14024)]?). A breakthrough result is due to \textit{J.-Y. Welschinger} [Invent. Math. 162, No. 1, 195--234 (2005; Zbl 1082.14052)] who used moduli spaces of real rational pseudo-holomorphic curves to define invariants. Among other names to be cited in the area are Itenberg, Kharlamov, Shustin, Eremenko, Gabrielov , Sottile and Soprunova. There exist connections to Gromov-Witten-Seiberg theory and estimates of real solutions to sparse polynomial systems.