Conformal TBA for resolved conifolds (Q2144557)

Counts of BPS states with fixed charge \(\gamma\) and asymptotic value \(z\) of moduli in \(4\) dimensions were rigorously interpreted by Bridgeland in terms of Donaldson-Thomas (DT) invariants. He showed that the complex hyper-Kähler metric on the holomorphic tangent bundle \(TS\) over the stability space \(S\) of their definition is determined by solutions to a suitable Riemann-Hilbert problem. In prior work, the authors of this paper reformulated the Riemann-Hilbert problem in terms of conformal TBA integral equations for the Darboux coordinates \(X_{\gamma}(\theta,z,t)\), where \(\theta\) parametrizes the fiber of \(TS\) and \(t\in\mathbb{C}^{\times}\) is a ``twistorial'' parameter of \(\mathbb{C}^{\times}\) action on \(S\). In this paper, the authors consider the case of an infinitely many uncoupled BPS structures, which holds for the derived categories of coherent sheaves on non-compact Calabi-Yau \(3\)-folds with no compact divisors. Small crepant resolutions of ADE singularities, like the resolved conifold, are primary examples. There, the BPS counts \(\Omega(\gamma,z)\) are non-zero on a finite set of `trajectories' \(\gamma=\beta+k\delta\), where \(\beta\) runs over homology classes of rational curves, \(\delta\) is the skyscraper sheaf (charge vector at a point), and \(k\) is an integer. The invariants \(n_{k,\beta}:=\Omega(\beta+k\delta,z)\) are commonly known in the literature as the Gopakumar-Vafa invariants, and they count BPS states of a D2-brane on a rational curve with \(k\) units of D0-brane charge. As always with uncoupled BPS structures, the TBA equations trivialize and (formally) produce explicit formulas for the Darboux coordinates, the Plebanski potential and the \(\tau\)-function. However, when there are infinitely many structures, the ``formulas'' are, in fact, divergent sums over \(\gamma\). Naively summing the finite case formulas over non-vanishing DT invariants produces one of the solutions found by Bridgeland, but only in the case of the resolved conifold and with \(\theta=0\). The paper proposes an alternative regularization procedure, where the formal sums are first rewritten in terms of rational DT invariants supported on \(\gamma=n(\beta+k\delta)\), and then the sum over \(k\) is regularized before summing over \(n\). The regularization of sums over \(k\) reduces to regularizing integrals of the form \(\int_0^\infty \frac{y^{v/w}}{1-y}\,\frac{dy}{y}\) at \(y=1\). This can be done in three ways, by moving the contour above the pole, below the pole, or taking the principal value. All three options produce well-defined solutions. However, the aforementioned Bridgeland's solution for the conifold is only recovered by choosing \textit{different} options for different types of classes \(\beta\), and the Darboux coordinates then have an essential singularity in \(t\) (contrary to Bridgeland's claim that they are meromorphic). On the other hand, consistently taking the principal value for all \(\beta\) produces better behaved coordinates, and recovers another of Bridgeland's solutions for the conifold with polynomial behavior at \(t\to\infty\). Moreover, the expansion of the \(\tau\)-function for small \(t\) reproduces the genus \(g\geq1\) terms of the topological string partition function, with the \(g=0\) contributions recoverable from the Plebanski potential. This opens up the prospect of finding a non-perturbative definition of the string partition function. Some calculations in the paper are based on novel integral representations for double and triple sine functions, usually defined via multiple Barnes gamma functions and closely related to quantum dilogarithms and trilogarithms. In particular, a representation for the triple sine function similar to the Woronowicz representation for Faddeev's quantum trilogarithm is stated as a conjecture and numerical evidence is presented in its favor.