Abstract
We extend a recent result of Dubrovin et al. in On tau-functions for the KdV hierarchy, arXiv:1812.08488 to the Toda lattice hierarchy. Namely, for an arbitrary solution to the Toda lattice hierarchy, we define a pair of wave functions and use them to give explicit formulae for the generating series of k-point correlation functions of the solution. Applications to computing GUE correlators and Gromov–Witten invariants of the Riemann sphere are under consideration.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The Toda lattice hierarchy, which contains the Toda lattice equation
is an important integrable hierarchy of nonlinear differential–difference equations [18, 19, 22, 27]. In this paper, following the idea of [13], we derive new formulae for generating series of k-point correlation functions for the Toda lattice hierarchy by using the matrix resolvent approach [10] and by introducing a pair of wave functions.
1.1 Toda lattice hierarchy and tau-function
Let
be the polynomial ring. Define the shift operator \(\Lambda :{{\mathcal {A}}}\rightarrow {{\mathcal {A}}}\) via
\(\forall \, i\in {{\mathbb {Z}}}\) and \(f,g\in {{\mathcal {A}}}\). Denote by \(\Lambda ^{-1}\) the inverse of \(\Lambda \) satisfying \(\Lambda ^{-1} (v_i) = v_{i-1}\), \(\Lambda ^{-1} (w_i) = w_{i-1}\), and \(\Lambda ^{-1} (f g) = \Lambda ^{-1} (f) \, \Lambda ^{-1} (g)\). For a difference operator P on \({{\mathcal {A}}}\), we mean an operator of the form \(P = \sum _{m\in {{\mathbb {Z}}}} P_m \, \Lambda ^m \), where \(P_m \in {{\mathcal {A}}}\). Denote \(P_+:=\sum _{m\ge 0} P_m \, \Lambda ^m\), \(P_-:=\sum _{m< 0} P_m \, \Lambda ^m\), \({\mathrm{Coef}}(P,m):=P_m\). A linear operator \(D:{{\mathcal {A}}}\rightarrow {{\mathcal {A}}}\) is called a derivation on \({{\mathcal {A}}}\), if
The derivation D is called admissible if it commutes with \(\Lambda \). Clearly, every admissible derivation D is uniquely determined by the values \(D(v_0)\) and \(D(w_0)\). Let
be a difference operator, and define a sequence of difference operators \(A_k\), \(k\ge 0\) by
We associate with \(A_k\) a sequence of admissible derivations \(D_k:{{\mathcal {A}}}\rightarrow {{\mathcal {A}}}\) defined via
The first few \(D_k(v_0)\) and \(D_k(w_0)\) are \(D_0(v_0)=w_1-w_0\), \(D_0(w_0)=w_0 \, (v_0-v_{-1})\); \(D_1(v_0)=w_1(v_1+v_0)-w_0(v_0+v_{-1})\), \(D_1(w_0)=w_0\bigl (w_1-w_{-1}+v_0^2-v_{-1}^2\bigr )\), etc.
Lemma 1
The operators \(D_k\), \(k\ge 0\) pairwise commute.
This lemma was known. We call \(D_k\) the Toda lattice derivations, and (5) the abstract Toda lattice hierarchy.
A tau-structure associated with the derivations \((D_k)_{k\ge 0}\) is a collection of polynomials \(\bigl (\Omega _{p,q}, S_p\bigr )_{p,q\ge 0}\) in \({\mathcal {A}}\) satisfying
for all \(p,q,r\ge 0\). It can be shown (e.g., [10]) that the tau-structure exists and is unique up to replacing \(\Omega _{p,q},S_p\) by \(\Omega _{p,q}+c_{p,q}\) and \(S_p+a_p\) respectively, where \(c_{p,q}=c_{q,p}\) and \(a_ p\) are arbitrary constants. The tau-structure \(\Omega _{p,q},S_p\) is called canonical if
Let us take \(\Omega _{p,q},S_p\) the canonical tau-structure. For \(m\ge 3\), define
By (6), we know that the \(\Omega _{p_1,\ldots ,p_m}\), \(m\ge 2\), are totally symmetric with respect to permutations of the indices \(p_1,\ldots ,p_m\). The first few of these polynomials are
If we think of \(v_0\), \(w_0\) as two functions v(n), w(n) of n, respectively, and \(v_i\), \(w_i\) as \(v(n+i)\), \(w(n+i)\), then the Toda lattice derivations \(D_k\) lead to a hierarchy of evolutionary differential–difference equations, called the Toda lattice hierarchy, given by
where \(k\ge 0\), and the \(D_k(v_0) (n), D_k(w_0) (n)\) are defined as \(D_k(v_0), D_k(w_0)\) with \(v_i\), \(w_i\) replaced by \(v(n+i)\), \(w(n+i)\), respectively. Lemma 1 implies that the flows (12) all commute. So we can solve the whole Toda lattice hierarchy (12) together, which yields solutions of the form \((v=v(n,\mathbf{t}),w=w(n,\mathbf{t}))\). Here, \(\mathbf{t}:=(t_0,t_1,\ldots )\) denotes the infinite time vector. Note that the \(k=0\) equations read
which are equivalent to Eq. (1) via the transformation
Here, dot, “ \(\dot{}\) ”, is identified with \(\partial /\partial t_0\).
Let V be a ring of functions of n closed under shifting n by \(\pm 1\). For two given \(f(n),g(n)\in V\), consider the initial value problem for (12) with the initial condition:
The solution \((v(n,\mathbf{t}), w(n,\mathbf{t}))\in V[[\mathbf{t}]]^2\) exists and is unique, which gives the following 1–1 correspondence:
Example 1
\( f(n)=0,~g(n)=n\). (For this case, one can take \(V={{\mathbb {Q}}}[n]\).) The corresponding unique solution governs the enumerations of ribbon graphs in all genera.
Example 2
\(f(n)= (n+\frac{1}{2})\epsilon , ~ g(n)=1\). (For this case, one can take \(V={{\mathbb {Q}}}[\epsilon ][n]\).) The corresponding unique solution governs the Gromov–Witten invariants of \({\mathbb {P}}^1\) in the stationary sector in all genera and all degrees.
Let \((v,w)\in V[[\mathbf{t}]]^2\) be an arbitrary solution to the Toda lattice hierarchy (12). Write \(\Omega _{p,q}(n,\mathbf{t})\) and \(S_p(n,\mathbf{t})\) as the images of \(\Omega _{p,q}\) and \(S_p\) under the substitutions
respectively. (Similar notations will be used for other elements of \({{\mathcal {A}}}\).) Equalities (6) then imply the existence of a function \(\tau =\tau (n,\mathbf{t})\) such that for \(p,q\ge 0\),
We call \(\tau (n,\mathbf{t})\) the Dubrovin–Zhang (DZ)-type tau-function [10, 15] of the solution (v, w), in short the tau-function of the solution. The symmetry in (9) is more obvious: the image \(\Omega _{p_1,\ldots ,p_m}(n,\mathbf{t})\) of \(\Omega _{p_1,\ldots ,p_m}\) under (16) satisfies
Define \(\Omega _p(n,\mathbf{t})=\partial _{t_p} \log \tau (n,\mathbf{t})\), \(p\ge 0\). These logarithmic derivatives of \(\tau (n,\mathbf{t})\) are called correlation functions of the solution (v, w). The specializations \(\Omega _{p_1,\ldots ,p_m}(n,\mathbf{0})\) are called m-point partial correlation functions of (v, w).
Remark 1
The tau-function \(\tau (n,\mathbf{t})\) of the solution (v, w) is unique up to multiplying it by the exponential of a linear function of \(n,t_0,t_1,t_2,\ldots \).
1.2 Matrix resolvent
The matrix resolvent (MR) method for computing correlation functions for integrable hierarchies was introduced in [1,2,3], and was extended to the discrete case in [10] (in particular to the Toda lattice hierarchy). Denote
The following lemma for the Toda lattice hierarchy was proven in [10].
Lemma 2
[10] There exists a unique series \(R(\lambda )\in {\mathrm{Mat}}\left( 2,{{\mathcal {A}}}\left[ \left[ \lambda ^{-1}\right] \right] \right) \) satisfying
The unique series \(R(\lambda )\) in Lemma 2 is called the basic matrix resolvent. The first few terms of \(R(\lambda )\) are given by
Proposition 1
[10] For any \(k\ge 2\), the following formula holds true:
where \({\mathcal {S}}_k\) denotes the symmetry group and \(C_k\) the cyclic group, and \(\pi (k+1):=\pi (1)\).
The meaning of (25) is the following: For any fixed permutation \((j_1,\ldots ,j_k)\) of \((1,\ldots ,k)\), expanding the right-hand side with respect to \(|\lambda _{j_1}|>\cdots>|\lambda _{j_k}|>>0\) gives identical formal power series with the left-hand side. This is because, after the summation over the \({\mathcal {S}}_k/C_k\) and subtracting \(\frac{\delta _{k,2}}{(\lambda _1-\lambda _2)^2}\), the poles in the diagonal cancel (cf. Proposition 2 of [12] for a straightforward proof of this point). We note that, as formal power series, the coefficients of the both sides of (25) are in \({{\mathcal {A}}}\). We give in Sect. 2 a new proof of (25), where we keep all derivations with coefficients in \({{\mathcal {A}}}\).
1.3 From wave functions to correlation functions
In [13], we introduced the notion of a tuple of wave functions (in many cases a pair) to the study of tau-function without using the Sato theory. Let us generalize it to the Toda lattice hierarchy. Our definition of a pair will be based on the standard construction of wave functions for the Toda lattice hierarchy [5, 6, 27]. For given (f(n), g(n)) a pair of arbitrary elements in V, let L be the linear difference operator \(L=\Lambda +f(n)+g(n) \Lambda ^{-1}\). Denote
The function s(n) is in a certain extension \({\widehat{V}}\) of V and is uniquely determined by \(\log g(n)\) up to a constant. Below we fix a choice of s(n). An element \(\psi _A(\lambda ,n)=\bigl (1+{\mathrm{O}}\left( \lambda ^{-1}\right) \bigr ) \,\lambda ^n\) in the module \({{\widetilde{V}}}\left[ \left[ \lambda ^{-1} \right] \right] \lambda ^n\) is called a (formal) wave function of type A associated with f(n), g(n), if \(L \bigl ( \psi _A(\lambda ,n)\bigr )= \lambda \psi _A(\lambda ,n)\). Here, \(\widetilde{V}\) is a ring of functions of n satisfying
An element \(\psi _B(\lambda ,n)=\bigl (1+{\mathrm{O}}\bigl (\lambda ^{-1}\bigr )\bigr ) \, e^{-s(n)} \lambda ^{-n}\) in the module \({\widetilde{V}} \left[ \left[ \lambda ^{-1} \right] \right] e^{-s(n)} \lambda ^{-n}\) is called a (formal) wave function of type B, if \(L\bigl ( \psi _B(\lambda ,n) \bigr ) = \lambda \psi _B(\lambda ,n) \). Let \( \psi _A \in {\widetilde{V}}\left[ \left[ \lambda ^{-1}\right] \right] \, \lambda ^n\) and \(\psi _B \in {\widetilde{V}} \left[ \left[ \lambda ^{-1} \right] \right] e^{-s(n)} \lambda ^{-n}\) be two wave functions of type A and of type B associated with (f(n), g(n)), respectively. Define
We call \(\psi _A,\psi _B\) form a pair if the following normalization condition holds:
The existence of a pair of wave functions is proved in Sect. 3.
Denote by \((v(n,\mathbf{t}),w(n,\mathbf{t}))\) the unique solution in \(V[[\mathbf{t}]]^2\) to the Toda lattice hierarchy with (f(n), g(n)) as its initial value, by \(\psi _A(\lambda ,n)\) and \(\psi _B(\lambda ,n)\) a pair of wave functions associated with (f(n), g(n)) and by \(\tau (n,\mathbf{t})\) the DZ-type tau-function of \((v(n,\mathbf{t}),w(n,\mathbf{t}))\). Introduce
Theorem 1
Fix \(k\ge 2\) being an integer. The generating series of k-point partial correlation functions has the following expression:
Theorem 1 gives an algorithm with the initial value (f(n), g(n)) as the only input for computing the \(k_{{\mathrm{th}}}\)-order logarithmic derivatives of the tau-function \(\tau (n,\mathbf{t})\) evaluated at \(\mathbf{t}=\mathbf{0}\) for \(k\ge 2\). Indeed, by solving the spectral problem \(L(\psi )=\lambda \psi \) with \(L=\Lambda +f(n)+g(n)\Lambda ^{-1}\) and with the normalization condition (28), one constructs a pair of wave functions; the coefficients in the \({\mathbf{t}}\)-expansion of \(\log \tau (n,\mathbf{t})\) are then obtained through algebraic manipulations by using (85). (Recall that in the inverse scattering method (cf., e.g., [18, 19]), an additional integral equation needs to be solved.) Two applications of Theorem 1 are given in Sect. 5. For a certain class of bispectral solutions (cf. [20]), it would be possible to give a canonical way of constructing a pair of wave functions, which was briefly mentioned in [13] for the KdV hierarchy; we plan to do this for KdV and for Toda lattice in a future publication.
Organization of the paper In Sect. 2, we review the MR method of studying tau-structure for the Toda lattice hierarchy. In Sect. 3, we prove the existence of a pair of wave functions. In Sect. 4, we prove Theorem 1 and several other theorems. Applications to the computations of GUE correlators and Gromov–Witten invariants of \({\mathbb {P}}^1\) are given in Sect. 5. In Appendix A, we give an extension of \({{\mathcal {A}}}\), define a pair of abstract pre-wave functions, and prove an abstract version for Theorem 1.
2 Matrix resolvent and tau-structure
We continue in this section with more details in reviewing the MR method [10] to the Toda lattice hierarchy. Denote by \({\mathcal {L}}\) the matrix Lax operator for the Toda lattice:
Let \(R(\lambda )\) be the basic matrix resolvent (of \({\mathcal {L}}\)). Write
where \(a_i,b_i,c_i\in {{\mathcal {A}}}\). From the defining Eqs. (21)–(23), we see that the series \(\alpha ,\beta ,\gamma \) satisfy the equations
These equalities give rise to the following recursion relation for \(a_i,b_i,c_i\):
along with
Equations (37)–(40) are called the matrix resolvent recursion relation.
It was proven [10] that the abstract Toda lattice hierarchy (5) can be equivalently written as
where \(j\ge 0\). Define an operator \(\nabla (\lambda )\) by
We have
Lemma 3
There exists a unique element \(W(\lambda ,\mu )\) in \({{\mathcal {A}}}\otimes {\mathrm{sl}}_2({{\mathbb {C}}}) \left[ \left[ \lambda ^{-1},\mu ^{-1}\right] \right] \lambda ^{-1} \mu ^{-1}\) of the form
satisfying the following linear inhomogeneous equations for the entries of W:
Proof
The existence part of this lemma follows from Lemma 2. Indeed, if we define
then \(W(\lambda ,\mu )\) satisfies (44)–(45). To see the uniqueness part, we first note that the (1,2)-entry and the (2,1)-entry of the matrix equation (44) imply that Y and Z can be uniquely expressed in terms of X. Indeed, we have
Substituting these two expressions in (45), we obtain the following linear inhomogeneous difference equation for X:
Suppose this equation has two solutions \(X_1,X_2\) in \({{\mathcal {A}}}\left[ \left[ \lambda ^{-1},\mu ^{-1}\right] \right] \lambda ^{-1} \mu ^{-1}\). Let \(X_0=X_1-X_2\), then \(X_0\in {{\mathcal {A}}}\left[ \left[ \lambda ^{-1},\mu ^{-1}\right] \right] \lambda ^{-1} \mu ^{-1}\), and it satisfies the following equation:
It follows that \(X_0\) vanishes. Indeed, write \(X_0=\sum _{j\ge 0} X_{0,j}(\mu ) \lambda ^{-(j+1)}\). Observe that
and recall that \(\alpha (\lambda ),\beta (\lambda ),\gamma (\lambda )\in {{\mathcal {A}}}\left[ \left[ \lambda ^{-1}\right] \right] \lambda ^{-1}\). Then, by comparing the coefficients of powers of \(\lambda ^{-1}\) consecutively, we find that \(X_{0,0}(\mu )=0\), \(X_{0,1}(\mu )=0\), \(X_{0,2}(\mu )=0\), \(\ldots \). So \(X_0=0\). Hence, \(X_1=X_2\). The lemma is proved. \(\square \)
Based on this lemma, we now give a new proof for the following proposition.
Proposition 2
[10] The following equation holds true:
where
Proof
Define \(W^*\) as the right-hand side of (50), i.e.,
More precisely, the entries of \(W^*\) have the expressions:
We can then verify that \(W^* \in {{\mathcal {A}}}\otimes {\mathrm{sl}}_2({{\mathbb {C}}}) \left[ \left[ \lambda ^{-1},\mu ^{-1}\right] \right] \lambda ^{-1} \mu ^{-1}\), as well as that \(W:=W^*\) satisfies Eqs. (44), (45). The latter is done by a lengthy but straightforward calculation. The proposition is proved due to Lemma 3. \(\square \)
If we define \({{\widetilde{\Omega }}}_{i,j}, {\widetilde{S}}_i\) by
then according to [10], \({{\widetilde{\Omega }}}_{i,j}, {\widetilde{S}}_i\) gives the canonical tau-structure for the Toda lattice, i.e.,
These equalities together with Proposition 2 lead to Proposition 1; see [10] for the detailed proof of Proposition 1.
Before ending this section, we will make two remarks. The first remark is that all the entries of \(R(\lambda )\) can be expressed by the canonical tau-structure. Indeed, we have
The proof was in [10]. The second remark is that existence of a tau-structure in general implies Lemma 1, and note that the proof in [10] of the fact that \({{\widetilde{\Omega }}}_{i,j}, {\widetilde{S}}_i\) is a tau-structure does not use the commutativity of the abstract Toda lattice hierarchy, so as a by-product of the matrix resolvent method we get a new proof of Lemma 1 together with a simple construction of the Toda lattice hierarchy. Similar idea was in [3].
3 Pair of wave functions
As in the Introduction, we start with the linear operator \(L(n)= \Lambda +f(n) + g(n) \, \Lambda ^{-1}\), where f(n) and g(n) are two given arbitrary elements in V. We show in this section the existence of pairs of wave functions associated with (f(n), g(n)). Let us write
Then, the spectral problems \(L(n) \bigl (\psi (\lambda ,n)\bigr ) = \lambda \psi (\lambda ,n)\) for \(\psi =\psi _A\) and for \(\psi =\psi _B\) recast into the following equations:
yielding recursions of the form (as equivalent conditions to (60)–(61))
where \(k\ge 0\). From these recursions, it easily follows that \(y_k,z_k\in V\), \(k\ge 0\). This proves the existence of wave functions of type A and of type B meeting the definitions in Sect. 1.3. Clearly, \(\psi _A\) and \(\psi _B\) are unique up to multiplying by arbitrary series \(G(\lambda )\) and \(E(\lambda )\) of \(\lambda ^{-1}\) with constant coefficient of the form \(G(\lambda )\in 1+ {{\mathbb {C}}}\left[ \left[ \lambda ^{-1}\right] \right] \lambda ^{-1}\) and \(E(\lambda ) \in 1+ {{\mathbb {C}}}\left[ \left[ \lambda ^{-1}\right] \right] \lambda ^{-1}\). Since \(\psi _A(\lambda ,n)=\bigl (1+{\mathrm{O}}(\lambda ^{-1})\bigr ) \,\lambda ^n\) and since \(\psi _B(\lambda ,n)=\left( 1+{\mathrm{O}}\left( \lambda ^{-1}\right) \right) \, e^{-s(n)} \lambda ^{-n}\), we find that the \(d(\lambda ,n)\) defined in (27) must have the form
Then, by using the definitions of wave functions and of s(n), one easily derives that
It follows that all \(d_k(n)\), \(k\ge 1\), are constants. Therefore, for any fixed choice of \(\psi _A\), we can suitably choose the factor \(E(\lambda )\) for \(\psi _B\) such that \(\psi _A,\psi _B\) form a pair. This proves the existence of pair of wave functions associated with f(n), g(n).
We proceed with the time dependence. Let \((v(n,\mathbf{t}),w(n,\mathbf{t}))\) be the unique solution in \(V[[\mathbf{t}]]^2\) to the Toda lattice hierarchy satisfying the initial condition \(v(n,\mathbf{0})=f(n)\), \(w(n,\mathbf{0})=g(n)\). Let \(L(n,\mathbf{t}) := \Lambda +v(n,\mathbf{t}) + w(n,\mathbf{t}) \, \Lambda ^{-1}\). Define \(\sigma (n,\mathbf{t})\) as the unique up to a constant function satisfying the following equations:
An element \(\psi _A(n,\mathbf{t},\lambda )= \bigl (1+\mathrm{O}\bigl (\lambda ^{-1}\bigr )\bigr ) \, \lambda ^n \, e^{\sum _{k\ge 0} t_k\lambda ^{k+1}}\) in \({\widetilde{V}}\left[ \left[ \mathbf{t}, \lambda ^{-1}\right] \right] \, \lambda ^n e^{\sum _{k\ge 0} t_k\lambda ^{k+1}}\) is called a wave function of type A associated with \((v(n,\mathbf{t}),w(n,\mathbf{t}))\) if
An element \(\psi _B(n,\mathbf{t},\lambda ) = \bigl (1+\mathrm{O}\bigl (\lambda ^{-1}\bigr )\bigr ) \lambda ^{-n} e^{-\sum _{k\ge 0} t_k\lambda ^{k+1}}\) in \({\widetilde{V}}\left[ \left[ \mathbf{t},\lambda ^{-1}\right] \right] e^{-\sigma (n,\mathbf{t})} \lambda ^{-n} e^{-\sum _{k\ge 0} t_k\lambda ^{k+1}}\) is called a wave function of type B associated with \((v(n,\mathbf{t}),w(n,\mathbf{t}))\) if
The existence of wave functions \(\psi _A\) and \(\psi _B\) of type A and of type B associated with \((v(n,\mathbf{t}),w(n,\mathbf{t}))\) is a standard result in the theory of integrable systems (cf. [5, 6, 13, 27]); therefore, we omit its details. Denote
and introduce
where \(Q(\mu ,n,\mathbf{t}) := -\frac{{\mathrm{id}}}{\mu } \; + \; \begin{pmatrix} 0 &{}\quad 0 \\ 0 &{}\quad \gamma (\mu ,n,\mathbf{t}) \\ \end{pmatrix}\). We know from, e.g., [10] that the wave function \(\psi _A(\lambda ,n,\mathbf{t})\) satisfies
Similarly, the wave function \(\psi _B(\lambda ,n,\mathbf{t})\) satisfies
Here, I denotes the \(2\times 2\) identity matrix.
Lemma 4
The following formula holds true:
Proof
Recalling definition (69) for d and using (71)–(72), we find
The lemma is then proved via a straightforward computation. \(\square \)
Definition 1
We say \(\psi _A,\psi _B\) form a pair if \( e^{\sigma (n-1,\mathbf{t})} d(\lambda ,n,\mathbf{t})=\lambda \).
The next lemma shows the existence of a pair.
Lemma 5
There exist a pair of wave functions \(\psi _A,\psi _B\) associated with \((v(n,\mathbf{t}),w(n,\mathbf{t}))\). Moreover, the freedom of the pair is characterized by a factor \(G(\lambda )\) via
with \(G_j\), \(j\ge 1\) being arbitrary constants.
Proof
Firstly, the freedom of a wave function \(\psi _A\) associated with (v, w) is characterized by the multiplication by a factor \(G(\lambda )\) of the form (76). Fix an arbitrary choice of \(\psi _A\). For \(\psi _B\) being a wave function of type B associated with (v, w), from (69) and the definitions of wave functions, we know \(e^{\sigma (n-1,\mathbf{t})}d(\lambda ,n,\mathbf{t})\) must have the form
for some \(d_k(n,\mathbf{t})\), \(k\ge 1\). By using (67), (68), (69), we find
i.e.,
Using Lemma 4 and (66), we have
So we have
We deduce from (77), (78), (79) that \(d_k(n,\mathbf{t})\), \(k\ge 1\), are all constants. Therefore, there exists a unique choice of \(\psi _B\) such that \(\psi _A,\psi _B\) form a pair. The lemma is proved. \(\square \)
4 The k-point generating series
Let \((v,w)=(v(n,\mathbf{t}),w(n,\mathbf{t}))\in V[[\mathbf{t}]]^2\) be the unique solution to the Toda lattice hierarchy with the initial value \((v(n,\mathbf{0}),w(n,\mathbf{0}))=(f(n),g(n))\), and \((\psi _A,\psi _B)\) a pair of wave functions associated with (v, w). Define
Proposition 3
The following identity holds true:
Proof
Define
It is easy to verify that M satisfies
The entries of M in terms of the pair of wave functions read
where we recall that \(d(\lambda ,n,\mathbf{t}) = \psi _A(\lambda ,n,\mathbf{t}) \, \psi _B(\lambda ,n-1,\mathbf{t}) - \psi _B(\lambda ,n,\mathbf{t}) \, \psi _A(\lambda ,n-1,\mathbf{t})\), which coincides with the determinant of \(\Psi (\lambda ,n,\mathbf{t})\). It follows from \(\psi _A(\lambda ,n,\mathbf{t})= \bigl (1+\mathrm{O}\bigl (\lambda ^{-1}\bigr )\bigr ) \, \lambda ^n \, e^{\sum _{k\ge 0} t_k\lambda ^{k+1}}\) and \(\psi _B(\lambda ,n,\mathbf{t}) = \bigl (1+\mathrm{O}\bigl (\lambda ^{-1}\bigr )\bigr ) \, e^{-\sigma (n,\mathbf{t})} \lambda ^{-n} \, e^{-\sum _{k\ge 0} t_k\lambda ^{k+1}}\) that
The proposition then follows from the uniqueness theorem proved in Sect. 2. \(\square \)
Define
Theorem 2
Fix \(k\ge 2\) being an integer. The generating series of k-point correlation functions of the solution \((v(n,\mathbf{t}),w(n,\mathbf{t}))\) has the following expression:
Proof
It follows from (81) that
where \(r_1(\lambda ,n,\mathbf{t}):=(\psi _A(\lambda ,n,\mathbf{t}), \psi _A(\lambda ,n-1,\mathbf{t}))\), \(r_2(\lambda ,n,\mathbf{t}):=(\psi _B(\lambda ,n-1,\mathbf{t}),-\psi _B(\lambda ,n,\mathbf{t}))\). Substituting this expression into the identity
we obtain
where we used definition (84) and
This proves the \(k=2\) case of (85). For \(k\ge 3\), the proof is similar. Indeed,
This proves the \(k\ge 3\) case of (85). The theorem is proved. \(\square \)
Remark 2
In (85) or (30), the freedom (75) affects the \(D(\lambda , \mu )\) through multiplying it by a factor of the form \(\frac{G(\lambda )}{G(\mu )}\), but the product \(\prod _{j=1}^k D(\lambda _{\pi (j)},\lambda _{\pi (j+1)})\) remains unchanged.
In Appendix A, the abstract form of (85) is obtained, where a pair of abstract pre-wave functions are introduced.
Proof of Theorem 1
Taking \(\mathbf{t}=\mathbf{0}\) on the both sides of (85) gives (30). \(\square \)
Write
Theorem 1 can then be alternatively written in terms of \(\phi _A,\phi _B\) by the following corollary.
Corollary 1
The following formula holds true for \(k\ge 2\):
where \(B(\lambda ,\mu ,n,\mathbf{t})\) is defined by
In particular, let \(\phi _A(\lambda ,n):=e^{(\Lambda -1)^{-1}(y(\lambda ,n))}\), \(\phi _B(\lambda ,n):=e^{(\Lambda -1)^{-1}(z(\lambda ,n))} e^{-s(n)}\) (cf. (60)–(61)), and let \(B(\lambda ,\mu ,n):= \frac{\phi _A(\lambda ,n) \, \phi _B(\mu ,n-1) \,-\, g(n)\, \phi _A(\lambda ,n-1) \,\phi _B(\mu ,n)}{\lambda -\mu }\), then we have
For some particular examples related to matrix models, it turns out that the suitable chosen D coincides, possibly up to simple factors, with certain kernel of the matrix model. However, the D is not unique. We now introduce a formal series \(K(\lambda ,\mu )\) such that the generating series of multi-point correlation functions still has an explicit expression, but this time K is local and is therefore unique for the given solution. The series K is defined by
where \(1+\alpha (\lambda )\) is the (1,1) entry of the basic matrix resolvent \(R(\lambda )\), and \(\gamma (\lambda )\) is the (2,1) entry. The next theorem expresses the left-hand side of (85) in terms of K.
Theorem 3
For any \(k\ge 2\), the following formula holds true:
Proof
The identity (81) gives
Substituting these expressions into (84), we obtain
where
Combining with the definition of \(K(\lambda ,\mu ,n,\mathbf{t})\) and Theorem 1, we find
The theorem is proved. \(\square \)
It seems to be an interesting question to study the geometric and algebraic meaning of the kernel K (as well as D). Below we give without proof some of their properties.
Proposition 4
The functions K and D are related to
We observe that the following three formal series
all belong to \({{\mathcal {A}}}\left[ \left[ \lambda ^{-1},\mu ^{-1}\right] \right] \). Therefore, the following three formal series
all belong to \(V[[\mathbf{t}]]\left[ \left[ \lambda ^{-1},\mu ^{-1}\right] \right] \). It follows from this observation and Proposition 4 that
Remark 3
We could loosen both the conditions for wave functions and the pair condition. Let us say \(\psi _A\) and \(\psi _B\) are pre-wave functions of type A and of type B, respectively, if they satisfy the first equations of (67) and (68). Define \(d_{{\mathrm{pre}}}(\lambda ,n,\mathbf{t})\) and \(D_{{\mathrm{pre}}}(\lambda ,\mu ,n,\mathbf{t})\) by (129) and (140). Then, the following formula holds true:
Now, \(\psi _A\) and \(\psi _B\) are determined by \((v(n,\mathbf{t}),w(n,\mathbf{t}) )\) up to
where \(G(\lambda ,\mathbf{t})=1+\sum _{k\ge 1} G_k(\mathbf{t}) \lambda ^{-k}\), \(E(\lambda ,\mathbf{t})=1+\sum _{k\ge 1} E_k(\mathbf{t})\lambda ^{-k}\) with \(G_k(\mathbf{t}), E_k(\mathbf{t}) \in {{\mathbb {C}}}[[\mathbf{t}]]\), \(k\ge 1\). This freedom affects \(D_{{\mathrm{pre}}}(\lambda ,\mu ,n,\mathbf{t})\) and \(d_{{\mathrm{pre}}}(\lambda ,n,\mathbf{t})\) into
Therefore, it gives rise to each summand of (100) the factor
which is equal to 1. Hence, the right-hand side of (100) remains unchanged.
5 Applications
Partition functions in some matrix models and enumerative models are particular tau-functions for the Toda lattice hierarchy. Theorem 1 can then be used for computing their logarithmic derivatives. In this section, we do two explicit computations.
5.1 Application I: enumeration of ribbon graphs
The initial data of the GUE solution to the Toda lattice hierarchy are given by \(f(n)=0\) and \(g(n)=n\); see, for example, [10] for the proof. For this case, we can take \(V={{\mathbb {Q}}}[n]\) and \({\widetilde{V}}=V\). Substituting the initial data in (26), we find
where C is a constant. Below, we fix this constant as 0.
Proposition 5
The \(\psi _A,\psi _B\) defined by
form a particular pair of wave functions associated with \((f(n)=0,g(n)=n)\). Here and below \((a)_{i}\) denotes the increasing Pochhammer symbol defined by \((a)_{i}=a(a+1)\ldots (a+i-1)\).
Proof
It is straightforward to verify that both \(\psi _A\) and \(\psi _B\) satisfy the equation
Moreover, from definitions (102)–(103), we see that
We are left to show that
Clearly, the meaning of this identity is the following: Both sides of (105) are Laurent series of \(\lambda ^{-1}\) with coefficients in \({\widetilde{V}}=V=Q[n]\), and the equality means all the coefficients should be equal. More precisely, the identity (105) can be equivalently written as the following sequence of identities:
From (64), we know that the left-hand side of (106) as a polynomial of n is a constant for any \(j\ge 0\). Note that the value of the left-hand side of (106) at \(n=0\) is obviously 0 for any \(j\ge 0\). The proposition is proved. \(\square \)
It follows from the above proposition an explicit expression for the \(D(\lambda ,\mu ,n,\mathbf{0})\) (cf. Eq. (84)) associated with the pair (102)–(103):
with \(A(\lambda ,\mu ,n)\) given by
This explicit expression (108) first appeared in [31]. Denote
As a corollary of Proposition 5, Theorem 1, and the above (107), we have achieved a new proof of the following theorem of Jian Zhou.
Theorem 4
[31] Fix \(k\ge 2\) being an integer. The generating series of k-point connected GUE correlators has the following expression:
where \({\widehat{A}}\) is defined by (108)–(109). Here, we recall that for any fixed \(i_1,\ldots ,i_k\), the connected GUE correlator \(\langle {\mathrm{tr}} \, M^{i_1}\ldots {\mathrm{tr}} \, M^{i_k}\rangle _{{\mathrm{c}}}\) is a polynomial of n (cf. [4, 10, 17, 21]).
5.2 Application II: Gromov–Witten invariants of \({\mathbb {P}}^1\) in the stationary sector
The initial data for the Gromov–Witten solution to the Toda lattice hierarchy were, for example, derived in [10,11,12]. It has the following explicit expression:
We have
where C is an arbitrary constant. Below, we take \(C=0\).
Proposition 6
The \(\psi _1,\psi _2\) defined by
form a particular pair of wave functions associated with \(f(n)=n\epsilon +\frac{\epsilon }{2},g(n)=1\). Here, \(J_\nu (y)\) denotes the Bessel function, and the right-hand sides of (112)–(113) are understood as the large \(\lambda \) asymptotics of the corresponding analytic functions.
Proof
Firstly, using the properties of Bessel functions, we can verify that \(\psi _A(\lambda ,n)\) and \(\psi _B(\lambda ,n)\) defined from the above asymptotics satisfy
Secondly, as \(\lambda \) goes to \(\infty \), the following asymptotics hold true:
Thirdly, \(\psi _A\) and \(\psi _B\) also satisfy
We have verified all the defining properties for a pair of wave functions associated with \(f(n)=n\epsilon +\frac{\epsilon }{2},g(n)=1\). The proposition is proved. \(\square \)
Note that
and denote
It follows from (84), (112)–(115) that the \(D(\lambda ,\mu ,0,\mathbf{0})\) associated with the pair (112)–(113) has the following explicit expression:
Then, according to [12], the function \(\frac{1}{\mu }D(\lambda ,\mu ,0,\mathbf{0})\) has the following expressions:
where \(a:=\frac{\mu }{\epsilon }\), \(b:=\frac{\lambda }{\epsilon }\), the \((a-b+1)_{-1}\) of (116) is defined as \(1/(a-b)\), and \(\sim \) in (118) is taken as \(a,b\rightarrow \infty \) away from the half integers. The explicit expression (118) first appeared in [12]. So we have completed a new proof of the following theorem.
Theorem 5
[12] The generating series of k-point (\(k\ge 2\)) Gromov–Witten invariants of \({\mathbb {P}}^1\) in the stationary sector has the following explicit expression:
where \({\widehat{A}} \, (\lambda ,\mu )\) is explicitly defined in (118), and
(See, for example, [12] for the notation about the integral in the right-hand side of (120).)
References
Bertola, M., Dubrovin, B., Yang, D.: Correlation functions of the KdV hierarchy and applications to intersection numbers over \(\overline{{\cal{M}}}_{g, n}\). Physica D 327, 30–57 (2016)
Bertola, M., Dubrovin, B., Yang, D.: Simple Lie algebras and topological ODEs. IMRN 2016, 1368–1410 (2018)
Bertola, M., Dubrovin, B., Yang, D.: Simple Lie algebras, Drinfeld–Sokolov hierarchies, and multi-point correlation functions. arXiv:1610.07534v2
Bessis, D., Itzykson, C., Zuber, J.B.: Quantum field theory techniques in graphical enumeration. Adv. Appl. Math. 1, 109–157 (1980)
Carlet, G.: The extended bigraded Toda hierarchy. J. Phys. A Math. Gen. 39, 9411–9435 (2006)
Carlet, G., Dubrovin, B., Zhang, Y.: The extended Toda hierarchy. Mosc. Math. J. 4, 313–332 (2004)
Di Francesco, P., Ginsparg, P., Zinn-Justin, J.: 2D gravity and random matrices. Phys. Rep. 254, 1–133 (1995)
Dickey, L.A.: Soliton Equations and Hamiltonian Systems, 2nd edn. World Scientific, Singapore (2003)
Dubrovin, B.: Geometry of 2D topological field theories. In: Francaviglia, M., Greco, S. (eds.) Integrable Systems and Quantum Groups (Montecatini Terme, 1993). Springer Lecture Notes in Mathematics, vol. 1620, pp. 120–348. Springer, Berlin, Heidelberg (1996)
Dubrovin, B., Yang, D.: Generating series for GUE correlators. Lett. Math. Phys. 107, 1971–2012 (2017)
Dubrovin, B., Yang, D.: On Gromov–Witten invariants of \({\mathbb{P}}^1\). Math. Res. Lett. 26, 729–748 (2019)
Dubrovin, B., Yang, D., Zagier, D.: Gromov–Witten invariants of the Riemann sphere. Pure Appl. Math. Q. (to appear)
Dubrovin, B., Yang, D., Zagier, D.: On tau-functions for the KdV hierarchy. arXiv:1812.08488
Dubrovin, B., Zhang, Y.: Virasoro symmetries of the extended Toda hierarchy. Commun. Math. Phys. 250, 161–193 (2004)
Dubrovin, B., Zhang, Y.: Normal forms of hierarchies of integrable PDEs, Frobenius manifolds and Gromov–Witten invariants. arXiv:math/0108160
Eguchi, T., Yang, S.-K.: The topological \(CP^1\) model and the large-\(N\) matrix integral. Mod. Phys. Lett. A 9, 2893–2902 (1994)
Harer, J., Zagier, D.: The Euler characteristic of the moduli space of curves. Invent. Math. 85, 457–485 (1986)
Faddeev, L.D., Takhtajan, L.A.: Hamiltonian Methods in the Theory of Solitons, Translated from Russian by Reyman, A.G. Springer, Berlin (1987)
Flaschka, H.: On the Toda lattice. II. Inverse-scattering solution. Prog. Theor. Phys. 51, 703–716 (1974)
Grünbaum, F.A., Yakimov, M.: Discrete bispectral Darboux transformations from Jacobi operators. Pac. J. Math. 204, 395–431 (2002)
Kazakov, V., Kostov, I., Nekrasov, N.: D-particles, matrix integrals and KP hierarchy. Nucl. Phys. B 557, 413–442 (1999)
Manakov, S.V., Complete integrability and stochastization of discrete dynamical systems. J. Exp. Theor. Phys. 67, 543–555 (in Russian) (English translation. In: Sov. Phys. JETP 40(2), 269–274 (1974))
Marchal, O.: WKB solutions of difference equations and reconstruction by the topological recursion. Nonlinearity 31, 226–262 (2017)
Mehta, M.L.: Random Matrices, 2nd edn. Academic Press, New York (1991)
Milanov, T.E.: Hirota quadratic equations for the extended Toda hierarchy. Duke Math. J. 138, 161–178 (2007)
Okounkov, A., Pandharipande, R.: Gromov–Witten theory, Hurwitz theory, and completed cycles. Ann. Math. 163, 517–560 (2006)
Ueno, K., Takasaki, K.: Toda lattice hierarchy. In: Okamoto, K. (ed.) Group Representations and Systems of Differential Equations (Tokyo, 1982), Advanced Studies in Pure Mathematics, vol. 4, pp. 1–95. North-Holland, Amsterdam (1984)
Watson, G.N.: A Treatise on the Theory of Bessel Functions, 2nd edn. Cambridge University Press, Cambridge (1944)
Zhang, Y.: On the \(CP^1\) topological sigma model and the Toda lattice hierarchy. J. Geom. Phys. 40, 215–232 (2002)
Zhou, J.: Emergent geometry and mirror symmetry of a point. arXiv:1507.01679
Zhou, J.: Hermitian one-matrix model and KP hierarchy. arXiv:1809.07951
Zhou, J.: Genus expansions of Hermitian one-matrix models: fat graphs vs. thin graphs. arXiv:1809.10870
Acknowledgements
The author is grateful to Youjin Zhang, Boris Dubrovin, Don Zagier for their advising and to Jian Zhou and Si-Qi Liu for helpful discussions. He thanks the referee for valuable suggestions; in particular, Appendix A comes out from the suggestions. He also wishes to thank Boris Dubrovin for introducing GUE to him and for helpful suggestions and discussions on this article. The work is partially supported by a starting research grant from University of Science and Technology of China.
Author information
Authors and Affiliations
Corresponding author
Additional information
Dedicated to the memory of Boris Anatol’evich Dubrovin, with gratitude and admiration.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A: Pair of abstract pre-wave functions
Appendix A: Pair of abstract pre-wave functions
Here, we construct a ring that is suitable for defining abstract pre-wave functions. Recall that \({{\mathcal {A}}}\) is the ring of polynomials of \(v_k,w_k\), \(k \in {{\mathbb {Z}}}\). Instead of the \({{\mathbb {Z}}}\)-coefficients, we will use in this appendix the \({{\mathbb {Q}}}\)-coefficients, i.e., \({{\mathcal {A}}}={{\mathbb {Q}}}\bigl [ \{v_k,w_k \,|\, k\in {{\mathbb {Z}}}\}\bigr ]\), is now under consideration. For each monic monomial \(\alpha \in {{\mathcal {A}}}\backslash {{\mathbb {Q}}}\), we associate a symbol \(m_\alpha \). Denote by \({\mathcal {B}}\) the polynomial ring
Define the action of \(\Lambda ^k\) on \({\mathcal {B}}\) with \(k\in {{\mathbb {Z}}}\) by
for \(\alpha _1,\ldots ,\alpha _l\) being monic monomials in \({{\mathcal {A}}}\backslash {{\mathbb {Q}}}\), as well as by linearly extending it to other elements of \({\mathcal {B}}\). For a monic monomial \(\alpha =v_{i_1} \ldots v_{i_r} w_{j_1} \ldots w_{j_s}\in {{\mathcal {A}}}\backslash {{\mathbb {Q}}}\) with \(i_1\le \cdots \le i_r\), \(j_1\le \cdots \le j_s\) and \(r+s\ge 1\), let \(k_\alpha :=-i_1\) (if \(r\ge 1\)), \(k_\alpha :=-j_1\) (if \(r=0\)); the monomial \(\Lambda ^{k_\alpha } (\alpha )\in {{\mathcal {A}}}\) is then called the (unique) reduced monomial (associated to \(\alpha \)). Denote by \({\mathcal {C}}\) the polynomial ring generated by \(m_{\beta }\), \(v_k\), \(w_k\) with \({{\mathbb {Q}}}\)-coefficients, where \(\beta \) are reduced monic monomials, and \(k\in {{\mathbb {Z}}}\). Let us also define an action of \(\Lambda ^k\) on \({\mathcal {C}}\), \(k\in {{\mathbb {Z}}}\). To this end, we introduce some notations: For \(\beta \) a reduced monic monomial of \({{\mathcal {A}}}\), denote
Then, for a monomial \(\alpha \cdot m_{\beta _1} \ldots m_{\beta _s}\) of \({\mathcal {C}}\) with \(\alpha \) being a monomial in \({{\mathcal {A}}}\), define
Define the action of \(\Lambda ^k\) on other elements in \({\mathcal {C}}\) by requiring it as a linear operator. Denote by \(p:{\mathcal {B}}\rightarrow {\mathcal {C}}\) the linear map which maps \(m_{\alpha _1} \ldots m_{\alpha _l}\in {\mathcal {B}}\) to \(n_{\alpha _1} \ldots n_{\alpha _l}\in {\mathcal {C}}\), for \(\alpha _i\), \(i=1,\ldots ,l\) being monic monomials in \({{\mathcal {A}}}\backslash {{\mathbb {Q}}}\). Denote by \({\mathcal {B}}^0\) the image of p. Clearly, \({{\mathcal {A}}}\subset {\mathcal {B}}^0\). Indeed, the element \((\Lambda -1) \bigl (\sum _{i=1}^l \lambda _i \, m_{\alpha _i}\bigr ) \in {\mathcal {B}}\) becomes \(\sum _{i=1}^l \lambda _i \alpha _i \in {{\mathcal {A}}}\) under the map p. Here, \(\alpha _1,\ldots ,\alpha _l\) are distinct monic monomials in \({{\mathcal {A}}}\backslash {{\mathbb {Q}}}\). Finally, we define an operator \({\mathbb {S}}: {{\mathcal {A}}}\backslash {{\mathbb {Q}}}\rightarrow {\mathcal {B}}^0\) by
for \(\alpha _1,\ldots ,\alpha _l\) being distinct monic monomials and \(\lambda _1,\ldots ,\lambda _l\in {{\mathbb {Q}}}\).
Motivated by (62) and (63), define two families of elements \(y_i,z_i \in {{\mathcal {A}}}\), \(i\ge 1\) by
Equivalently, the generating series \(y(\lambda ):=\sum _{i\ge 1} y_i/\lambda ^i\), \(z(\lambda ):=\sum _{i\ge 1} z_i/\lambda ^i\) satisfy
Define
where \(e^{-\sigma }\) is a formal element satisfying \(e^{(1-\Lambda ^{-1})(-\sigma )} = w_0\), and \(\lambda ^n\), \(\lambda ^{-n}\) are formal elements satisfying \(\Lambda ^k (1\otimes \lambda ^n)=\lambda ^k \otimes \lambda ^n\), \(\Lambda ^k (1\otimes \lambda ^{-n})=\lambda ^{-k} \otimes \lambda ^{-n}\), \(k\in {{\mathbb {Z}}}\). We have
where \(L=\Lambda +v_0 + w_0 \, \Lambda ^{-1}\). We call \(\psi _A\) and \(\psi _B\) the abstract pre-wave functions of type A and of type B, respectively, associated with \(v_0,w_0\).
Denote
and
Then, we have the following identity:
The proof is similar to that of Proposition 3. (The main fact used in the proof is that from the definition, the coefficients of entries of \(R(\lambda )\) are uniquely determined in an algebraic way.) We omit its details here. However, let us explain the equality (131) by an equivalent form. From definition, we have
Then, from a straightforward calculation by using the definitions, we find
Hence, the equality (131) means new expressions for the entries of the basic matrix resolvent \(R(\lambda )\) explicitly in terms of \(y(\lambda ),z(\lambda )\). Substituting the following expansions
into (132)–(135), we find that the new expressions agree with (24). Combining with (56), (57), we obtain
We therefore arrive at the following formulae:
Let us proceed to the generating series of multi-point correlation functions. Define
Using (131), Proposition 1, and a similar argument to the proof of Theorem 2, we obtain
For the reader’s convenience, we give the first few terms of the abstract pre-wave functions \(\psi _A(\lambda )\) and \(\psi _B(\lambda )\) as follows:
It turns out that the above abstract pre-wave functions form a pair. Namely, \(d_{{\mathrm{pre}}}(\lambda ) = \lambda \, e^{\Lambda ^{-1}(-\sigma )}\). Interestingly, for given arbitrary initial value (f(n), g(n)), based on this statement, one obtains a constructive method for a pair of wave functions associated with (f(n), g(n)) (cf. (28) in Sect. 1.3 for the definition of a pair). This is important considering Theorem 1. We hope to confirm the statement on the pair property of the abstract pre-wave functions in another publication.
Rights and permissions
About this article
Cite this article
Yang, D. On tau-functions for the Toda lattice hierarchy. Lett Math Phys 110, 555–583 (2020). https://doi.org/10.1007/s11005-019-01232-5
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11005-019-01232-5