A topological view of Gromov-Witten theory (Q2503376)

A fruitful guiding principle in Gromov-Witten theory is that, since quantum cohomology is a deformation of ordinary cohomology, each classical topology cohomological construction should have a Gromov-Witten analogue. In the present paper, the authors investigate a Gromov-Witten analogue of the long exact sequence for the cohomology of a pair, obtaining the following remarkable result: given a nonsingular projective variety \(V\) containing a nonsingular divisor \(W\), the relative Gromov-Witten theory of the pair \((V,W)\) is uniquely and effectively reconstructed from the absolute Gromov-Witten theories of \(V\) and \(W\) and from the restriction map \(H^*(V;{\mathbb Q})\to H^*(W;{\mathbb Q})\). As a corollary, the authors are able to determine a complete mathematical presentation of the Gromov-Witten theory (in all genera) of the Calabi-Yau quintic hypersurface in \({\mathbb P}^4\). The proof of the theorem goes as follows. First, one considers a line bundle \(N\) over a nonsingular projective variety \(W\). There is a natural \({\mathbb C}^*\)-action on the fibers of the projective bundle \({\mathbb P}(N\oplus {\mathcal O}_W)\to W\), with fixed loci the divisors \(D_0\) and \(D_\infty\) corresponding to the summands \(N\) and \({\mathcal O}_W\) respectively; both divisors are canonically isomorphic to \(W\). The virtual localization formula then computes the absolute Gromov-Witten theory of \({\mathbb P}(N\oplus {\mathcal O}_W)\) entirely in terms of the Gromov-Witten theory of \(W\). Next, the authors show that also the relative theories of the three pairs \(({\mathbb P}(N\oplus {\mathcal O}_W), D_0)\), \(({\mathbb P}(N\oplus {\mathcal O}_W), D_\infty)\), and \(({\mathbb P}(N\oplus {\mathcal O}_W), D_0\cup D_\infty)\) can be uniquely and effectively reconstructed from the Gromov-Witten theory of \(W\) and the first Chern class of \(N\). This can be seen as a Leray-Hirsch result in Gromov-Witten theory; its proof is far from being trivial and uses localization and degeneration in order to obtain an explicit set of recursions which determine the relative theories of the three pairs. The next step in the proof of the main result of the paper is now clear: one uses the above construction as a description of `Gromov-Witten theory of a tubular neighborhood' of a smooth divisor \(W\) in a smooth variety \(V\). More precisely, let \(N\to W\) be the normal bundle of \(W\) in \(V\), and let \(\mathcal F\) be the degeneration to the normal cone of \(W\), i.e., the blow-up of \(V\times {\mathbb C}\) along the subvariety \(W\times \{0\}\). If \(\varepsilon:{\mathcal F}\to {\mathbb C}\) is the projection to the second factor, one has \(\varepsilon^{-1}(0)=V\cup_W {\mathbb P}(N\oplus {\mathcal O}_W)\), where \(W\hookrightarrow {\mathbb P}(N\oplus {\mathcal O}_W)\) is determined by the identification of \(W\) with the divisor \(D_0\) corresponding to the summand \(N\). The degeneration formula then expresses the absolute Gromov-Witten theory of \(V\) in terms of the relative theories of the two pairs \((V,W)\) and \(({\mathbb P}(N\oplus {\mathcal O}_W),W)\) [\textit{Y.~Eliashberg}, \textit{A.~Givental} and \textit{H.~Hofer}, GAFA 2000, 560--673 (2000; Zbl 0989.81114); \textit{E.-N.~Ionel} and {T.~H.~Parker}, Ann. Math. (2) 157, No. 1, 45--96 (2003; Zbl 1039.53101); \textit{A.-M.~Li} and \textit{Y.~Ruan}, Invent. Math. 145, No. 1, 151--218 (2001; Zbl 1062.53073); \textit{J.~Li}, J. Differ. Geom. 60, No. 2, 199--293 (2002; Zbl 1063.14069)]. By the Leray-Hirsch-type theorem described above, the relative theory of \(({\mathbb P}(N\oplus {\mathcal O}_W),W)\) is enterely determined by the absolute theory of \(W\) and by the class \(c_1(N)=\iota^*(c_1(T_V))-c_1(T_W)\), where \(\iota^*:H^*(V;{\mathbb Q})\to H^*(W;{\mathbb Q})\) is the restriction map. Summing up, via the degeneration formula, the absolute theory of \(V\) is determined by the absolute theory of \(W\), the restriction map \(\iota^*\), and the relative theory of \((V,W)\). The authors are able to invert the degeneration formula, thus proving their main result. As a corollary, they obtain the following Mayer-Vietoris result in Gromov-Witten theory: let \(V\) be a nonsingular projective variety and let \(\varepsilon:{\mathcal V}\to \Delta\) be a flat family over a disk at the origin such that \({\mathcal V}\) is nonsingular, \(\varepsilon\) is smooth over \(\Delta\setminus\{0\}\), \(\varepsilon^{-1}(1)\simeq V\) and \(\varepsilon^{-1}(0)=V_1\cup_W V_2\) is a normal crossing divisor in \({\mathcal V}\); then the image of \(H^*(V_1\cup_W V_2;{\mathbb Q})\) in \(H^*(V;{\mathbb Q})\) determined by the map \(\varepsilon\) can be uniquely and effectively reconstructed from the absolute theories of \(V_1\), \(V_2\) and \(W\) and from the restriction maps \(H^*(V_1;{\mathbb Q})\to H^*(W;{\mathbb Q})\) and \(H^*(V_2;{\mathbb Q})\to H^*(W;{\mathbb Q})\). Another closely related result is the following blow-up lemma: if \(Z\hookrightarrow V\) is the nonsingular complete intersection of two nonsingular divisors \(W_1\) and \(W_2\) of the nonsingular projective variety \(V\), then the Gromov-Witten theory of the blow-up \(\widetilde{V}\) of \(V\) along \(Z\) is uniquely and effectively determined by the Gromov-Witten theories of \(V\), \(W_1\) and \(Z\) and by the restriction maps \(H^*(V;{\mathbb Q})\to H^*(W_1;{\mathbb Q})\to H^*(Z;{\mathbb Q})\); note that the Gromov-Witten theory of only one of the two divisors \(W_1\), \(W_2\) is needed. In the final part of the paper, the authors show how the above results effectively determine the full (i.e., in all genera) Gromov-Witten theory of the quintic threefold in \({\mathbb P}^4\) in terms of the Gromov-Witten theories of the following spaces: \({\mathbb P}^3\), \({\mathbb P}^2\), \(S_2\), \(S_3\), \(C_{1,2}\), \(C_{,3}\), \(C_3,4\) and \(C_{4,5}\), where \(S_d\) is a nonsingular degree \(d\) surface in \({\mathbb P}^3\) and \(C_{d_1,d_2}\) is a nonsingular complete intersection curve of type \((d_1,d_2)\) in \({\mathbb P}^3\); the Gromov-Witten theories of these spaces are all known [\textit{J.~Bryan} and \textit{N.~C.~Leung}, J. Am. Math. Soc. 13, No.2, 371--410 (2000; Zbl 0963.14031); \textit{A.~B.~Givental}, Int. Math. Res. Not. 2001, No. 23, 1265--1286 (2001; Zbl 1074.14532); \textit{T.~Graber} and \textit{R.~Pandharipande}, Invent. Math. 135, No.2, 487--518 (1999; Zbl 0953.14035); \textit{A.~Okounkov} and \textit{R.~Pandharipande}, Ann. Math. (2) 163, No. 2, 517--560 and 561--605 (2006; Zbl 1105.14076) and Ann. Math. (2) 163, No. 2, 561-605 (2006; Zbl 1105.14077), Invent. Math. 163, No. 1, 47--108 (2006; Zbl 1140.14047)]. The effective computation of the Gromov-Witten theory of the quintic is, however, extremely difficult. The authors remark that the results contained in the present paper could be used to extend Gathmann's computations from genera 0 and 1 to all genera, obtaining a determination of the Gromov-Witten theory of the quintic threefold which is likely to be more suitable for explicit computations [\textit{A.~Gathmann}, Duke Math. J. 115, No. 2, 171--203 (2002; Zbl 1042.14032), Math. Ann. 325, No. 2, 393--412 (2003; Zbl 1043.14016)].