Mean field equations, hyperelliptic curves and modular forms. II (Q2011992)

The paper under review is a sequel to the authors' paper [\textit{C.-L. Chai}, \textit{C.-S. Lin} and \textit{C.-L. Wang}, Camb. J. Math. 3, No. 1--2, 127--274 (2015; Zbl 1327.35116)] in which they ``introduce a \textit{pre modular form} \(Z_n(\sigma;\tau)\) of weight \(\frac{1}{2}n(n+1)\) \(\forall n \in \mathbb{N}\), where \((\sigma,\tau) \in \mathbb{C} \times \mathbb{H}\), such that for \(E_\tau=\mathbb{C}/(\mathbb{Z}+\mathbb{Z}_\tau)\), every non-trivial zero of \(Z_n(\sigma;\tau)\) (i.e., \(\sigma \notin E_\tau[2]\)) corresponds to a (scaling family of) solution to the singular Liouville (mean field) equation (MFE) \[ \Delta u+e^u=\rho \delta_0 \leqno{(1)} \] on the flat torus \(E_\tau\), where \(\Delta = \frac{\partial^2}{\partial x^2} + \frac{\partial^2}{\partial y^2}\) denotes the Laplace operator on \(E\) induced from \(\mathbb{C}\), \(n \in \mathbb{N}\) and \(\delta_0\) is the Dirac measure at \([0] \in E\) and \(\rho = 8\pi n\).'' The study of (1) was initiated by the authors in their earlier paper (see equation (1.2) p. 912 of [\textit{C.-S. Lin} and \textit{C.-L. Wang}, Ann. Math. (2) 172, No. 2, 911--954 (2010; Zbl 1207.35011)]) and further extended and developed by them in the first part of this paper (see equation (0.1.3) p. 129 of Zbl 1327.35116). It is known that the solvability of (1) depends on the moduli \(\tau \) in an intricate manner, this question, for the case \(n=1\), was settled by the authors themselves in their papers [loc. cit.; \textit{C.-S. Lin} and \textit{C.-L. Wang}, ``On the minimality of extra critical points of Green functions on flat tori'', Int. Math. Res. Not., doi:10.1093/imrn/rnw176 (2016); \textit{Z. Chen} et al., ``Green function, Painlevé VI equation, and Eisenstein series of weight one'', J. Differ. Geom. (to appear)]. In the paper under review the authors develop a theory with the help of modular forms to study the complete dependence of the solution of (1) \(\forall n \in \mathbb{N}\) and thereby they fully settle the question of solvability of (1). The authors define a \textit{pre-modular form} \(h(z;\tau)\) in Definition 0.1 as below: {Definition 0.1:} An analytic function \(h\) in \((z,\tau) \in \mathbb{C} \times \mathbb{H}\) is \textit{pre-modular} of weight \(k \in \mathbb{N}\) if it satisfies (i) For any fixed \(\tau\), the function \(h(z)\) is analytic in \(z\) and \(\overline{z}\) and it depends only on \(z(\bmod{\Lambda_\tau}) \in E_\tau\). (ii) For any fixed \textit{torsion type} \(z(\bmod{\Lambda_\tau}) \in E_\tau[N]\), the function \(h(\tau)\) is modular of weight \(k\) with respect to \(\Gamma (N)\), where, \(\Gamma (N) = \{ A \in \mathrm{SL}(2,\mathbb{Z})\mid A \equiv I_2(\bmod{N}) \}\). They further mention that they have exploited the concept of pre-modular forms ``to study \textit{deformations} in \(z\) to relate \textit{different} modular forms corresponding to different torsion points'' in their paper [\textit{Z. Chen} et al., loc. cit.] ``to achieve a complete solution to (1) for \(n = 1\) and for all \(\tau \).'' The present paper advances the study of the authors [\textit{Z. Chen} et al., loc. cit.] of finding a general solution of (1) from the pre-modular form \(Z_1 = Z\) to a general \(Z_n\), \(\forall n \geqslant 2\). By considering the rational meromorphic function \[ \mathbf{z}_n(a): = \zeta (\sum_{i=1}^n a_i)-\sum_{i=1}^n \zeta(a_i) \leqno{(2)} \] on \(E^n\) and defining the map \[ \sigma_n:{\overline X_n} \to E,\;[a] \mapsto \sigma_n([a]): = \sum_{i=1}^n [a_i] \leqno{(3)} \] (where, \([a]: = a(\bmod{\Lambda}) = \{ [a_1], \ldots,[a_n]\} \in X_n(\tau)\)) the authors show Theorem 0.2 (= Theorem 2.4): The map \(\sigma_n:{\overline X}_n \to E\) has degree \(\frac{1}{2}n(n+1)\). This theorem ensures the existence of a polynomial \(W_n(z) \in \mathbb{Q}[g_2,g_3,\wp (\sigma),\wp'(\sigma)][\mathbf{z}]\) of degree \(\frac{1}{2}n(n+1)\) in \(\mathbf{z}\) (where, \(\wp\) is the Weierstrass \(\wp\)-function) which defines the (branched) covering map \(\sigma_n\) for which the following result holds: Theorem 0.3 (= Theorem 3.2): The rational function \(\mathbf{z}_n \in K({\overline X_n})\) is a primitive generator for the field extension \(K({\overline X_n})\) over \(K(E)\) which is integral over the affine curve \(E^{\times}(:= E\backslash \{[0]\})\). This theorem has been proved in Section 3 of the paper using the concept of the \textit{tensor product} of two Lamé equations. The exact construction of the desired pre-modular form, which is the crux of the paper, is given by the authors as below: \[ Z_n(\sigma;\tau):=W_n(Z)(\sigma;\tau) \leqno{(4)} \] for which the following notable result (when \(n \geqslant 1\)) is established in Section 4 of the paper: Theorem 0.4: To every scaling family \(\{u_\lambda\}\) of solutions to the singular Liouville equation (1) on \({E_\tau}\), the zero set \(a(=(a_1, \ldots,a_n) \in \mathbb{C}^n) \in X_n\) of its normalized developing map \(f\) satisfies \(Z_n(\sigma_n(a);\tau)=0\) with \(\sigma_n(a) \notin E_\tau[2]\). Conversely, given \(\sigma_0 \in E_\tau\backslash E_\tau[2]\) with \(Z_n(\sigma_0;\tau)=0\), there is a unique \(a \in X_n\) with \(\sigma_n(a) = \sigma_0\) and it determines a developing map \(f =w_a/w_{-a}\) of a scaling family of solutions to equation (1). Here \(w_a\) and \(w_{-a}\) are are independent solutions to the integral Lamé equations on \(E\) \[ w'' = (n(n+1)\wp + B)w \leqno{(5)} \] with the spectral parameter \(B\). Further, it is to be noted at this point that for \(\sigma \in E_\tau[N]\), the \(N\)-torsion points, the modular forms \(Z_2(\sigma;\tau)\) and \(Z_3(\sigma;\tau)\) were first constructed by \textit{S. Dahmen} in [Counting integral Lamé equations with finite monodromy by means of modular forms. Master thesis. Utrecht: Utrecht University (2003)] wherein he had also conjectured the the existence of a modular form \(Z_n(\sigma;\tau)\) of weight \(\frac{1}{2}n(n+1)\) for \(n \geqslant 4\). This conjecture of Dahmen has been proved by the authors' in this paper by means of the results established in the theorems stated above. Since the addition map \(\sigma_n\) of the authors given in (3) above, being purely algebraic in nature, allows one to work out the explicit form of the polynomial \(W_n(\mathbf{z})\) for any \(n \in \mathbb{N}\) hence, \(Z_n\), but the process of this computation being often very tedious and cumbersome besides being time-consuming, the authors have, in the pursuit of a more efficient method of computation of \(Z_n\) turned their attention in a different direction and have shown that their addition map \(\sigma_n\) of (3) agrees with the projection map \(\pi_n:{\overline X_n} \to E\) introduced by \textit{R. S. Maier} in [Philos. Trans. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 366, No. 1867, 1115--1153 (2008; Zbl 1153.37425)] by appealing to the \textit{finite band integration theory} of the Lamé curve. While developing the main body of the paper, the authors inquire into the geometry of the hyperelliptic projection \(B:{\overline X_n} \to \mathbb{P}^1(\mathbb{C})\) in the first section of the paper where they prove that by using the \textit{Hermite-Halphen ansatz} solution of (5) ``solving equation (1) is equivalent to solving the the integral Lamé equation (5) with \textit{unitary projective monodromy groups}''. The geometry of the map \(\sigma_n:{\overline X_n} \to E\) is discussed in the second section, wherein the authors, after stating a basic proposition, prove the Theorem 0.2 mentioned supra by appealing to the principle of the \textit{theorem of the cube} given in [\textit{D. Mumford}, Abelian varieties. Cambridge: Oxford University Press (1974; Zbl 0326.14012)]. A proof of the Theorem 0.3 is presented in Section 3 of the paper, where the authors state this theorem in a more exact form as follows: Theorem 3.2: There is a weighted homogeneous polynomial \[ W_n(\mathbf{z}) \in \mathbb{Q}[g_2,g_3,\wp (\sigma),\wp '(\sigma)][\mathbf{z}] \] of \(\mathbf{z}\)-degree \(d_n= \deg \sigma_n\) such that for \(\sigma = \sigma_n(a) = \sum a_i\), we have \[ W_n(\mathbf{z}_n)(a)=0. \] Here, the weights of \(\mathbf{z},\wp (\sigma),\wp '(\sigma),g_2,g_3\) are 1, 2, 3, 4, 6 respectively. Indeed, \(\mathbf{z}_n(a)\) is a primitive generator of the finite extension of rational function field \(K({\overline X_n})\) over \(K(E)\) with \(W_n(\mathbf{z})\) being its minimal polynomial. Moreover, the extension is integral over the affine curve \(E^{\times}\). In the course of proving this theorem they state another important result in the form of Theorem 3.3 as below: Theorem 3.3: Let \(a,b \in Y_n\) and \((a_1, \ldots,a_n),(b_1, \ldots,b_n) \in \mathbb{C}^n\) be representatives of \(a,b\) such that \[ \sum_{i = 1}^n a_i=\sum_{i = 1}^n b_i,\;\sum_{i = 1}^n {\zeta (a_i)} = \sum_{i = 1}^n \zeta (b_i). \leqno{(6)} \] Suppose that \(\sum \wp (a_i) \neq \sum \wp (b_i)\). Then \(a,b\) are branch points of \({Y_n} \to \mathbb{C}\) which contains the same number of half periods. Equivalently, the Lamé functions \({w_a},{w_b}\) are of the same type. The authors present two proofs for Theorem 3.3 in the Sections 3.2 and 3.3 of the paper. Both these proofs of Theorem 3.3 are in turn based on a lemma stated, which is based on the tensor product of two Lamé equations. The lengthy first proof of this theorem is based on the statements and proofs of three lemmas and one proposition. That, the degree of the rational function \(\mathbf{z}_n\) equals the degree of \(\sigma_n\), viz. \(\frac{1}{2}n(n+1)\) is proved in Theorem 3.13. The fourth section of the paper is devoted to the completion of the proof of the Theorem 0.4 wherein it is also extended ``in terms of \textit{monodromy groups of Lamé equations}'' in the Section 4.2. Drawing from the work of Maier [loc. cit.] the authors present a computation of \(Z_n\) in the fifth section of the paper, where they use the \textit{Hermite-Krichever ansatz} for the solution of (5) in terms of the \textit{Lamé, twisted Lamé and the theta-twisted Lamé polynomials} employing the notations which are slightly different from those used in Maier's work. In Theorem 5.6 they establish that the Maier's projection map \(\pi_n:{\overline X_n} \to E\) agrees with their addition map \(\sigma_n:{\overline X_n} \to E\). In Section 5.2 the authors give explicit constructions of \(Z_n (n\leq4)\) using a result of Maier [loc. cit.] (which is stated as Theorem 5.3 in this paper) and Theorem 5.6 where, they reconstruct the polynomials \(Z_2,Z_3\) of Dahmen [loc. cit.] in Examples 5.8 and 5.9 and construct their weight 10 pre-modular form \(Z_4 (n=4)\) in example 5.10 using \textit{Mathematica}. They point out that whereas the computation of \(Z_3\) using \textit{Mathematica} took only a few seconds but the \textit{Mathematica} computation of \(Z_4\) required ``a couple hours''! The Remark 5.14 of the authors where they point out that in Maier's work the formula `` \(\deg {\pi_n} = \frac{1}{2}n(n+1) = \deg \kappa\) which was quoted as a significant formula from \textit{finite-band integration theory} ... we want to point out that it also follows from Theorem 5.6 and our degree formula \(\deg \sigma_n= \frac{1}{2}n(n+1) = \deg {z_n}\) in Theorem 2.4 and Theorem 3.13'' has the present reviewer's approbation. Finally, the appendix written by Y. C. Chou which presents a systematic confirmation of the Dahmen's conjectural counting formula for integral Lamé equations with finite monodromy for the case \(n=4\) by using the theory of pre-modular forms developed in Sections 4 and 5 of the paper is an important adjuvant to the paper. Towards the conclusion, the reviewer would like to mention that, if the word `hypergeometric' appearing twice in the lines of text of the paper just above the equation (1.9) on p. 567 be replaced by the words `confluent hypergeometric' at both these places then it would render the meaning of this statement more clear and precise.