Large \(N\) duality, Lagrangian cycles, and algebraic knots (Q1950366)

This paper marks a major advance in large \(N\) duality between the Chern-Simons theory on \(S^3\) and the Gromov-Witten theory on the resolved conifold. Witten showed that the Chern-Simons theory on a \(3\)-manifold \(M\) can be interpreted as a theory of (degenerate) holomorphic curves on the cotangent bundle \(T^*M\) represented by ribbon graphs in \(M\). For \(M=S^3\) Gopakumar and Vafa conjectured that as \(T^*S^3\) undergoes a geometric transition into the resolved conifold (the \(\mathcal{O}(-1)\oplus\mathcal{O}(-1)\) bundle over \(\mathbb{C}P^1\)) this degenerate theory is transformed into an equivalent theory of genuine holomorphic curves there. Their conjecture, named large \(N\) duality, has since been confirmed directly, for the partition functions and for the Wilson loop expectation values of the unknot, and indirectly, e.g. by checking its integrality predictions for the coefficients of HOMFLY polynomials of arbitrary links. In this paper the authors extend the direct verification to torus knots, and an indirect one to algebraic knots, by deriving for them from large \(N\) duality a conjecture of Oblomkov-Shende that relates BPS states from Donaldson-Thomas like invariants to HOMFLY polynomials. The underlying advance is a novel general construction of Lagrangian submanifolds in the resolved conifold that correspond to algebraic knots in \(S^3\), which enables direct localization computations in the case of torus knots. Given a knot \(K\) in \(S^3\) it is expected that the holomorphic curves of the dual theory on the resolved conifold are the open ones with boundaries on a Lagrangian submanifold \(L_K\). This \(L_K\) should be ``related'' to the conormal bundle \(N_K^*\subset T^*S^3\) via the conifold transition. However, \(N_K^*\) intersects the zero section, which shrinks into a singular point under the conifold transition which is then resolved. And naively taking \(N_K^*\) through the transition ``relates'' it to a singular set, which is neither Lagrangian nor a submanifold, except when \(K\) is the unknot. This issue plagued the theory for years preventing computations of open Gromov-Witten invariants in other examples. Mariño and Vafa suggested early on that \(N_K^*\) should first be deformed away from the zero section, and only then put through the transition. A construction along these lines was carried out by the reviewer using symplectic methods, however the resulting \(L_K\) were only totally real, and lacked symmetries needed for computations. In this paper the authors suggest a more general type of deformation, which in the case of algebraic knots results in explicitly describable Lagrangian submanifolds. For torus knots these \(L_K\) allow formal virtual localization computations analogous to the ones for the unknot (rigorous theory of open Gromov-Witten invariants is still unavailable), which confirm the duality. The proof is an open Gromov-Witten reflection of the Chern-Simons \(S\)-matrix formula that relates HOMFLY polynomials of torus knots to the colored invariants of the unknot. Moreover, rigid holomorphic cylinders that connect the zero section in \(T^*S^3\) to the deformed \(N_K^*\) are constructed, and they are shown to be transformed by the conifold transition into singular holomorphic disks with the boundary on \(L_K\). This completes the geometric picture of the duality, and opens a path towards proving it by deformation arguments.