A Sugawara–Legendre mechanism
for the four-point Heisenberg algebraThanks: Supported by FAPESP grant 2024/14914-9.

Fix an integer $m\geq 2$, the covering degree, and a polynomial $p(t)\in\mathbb{C}[t]$ of degree $d$ with no multiple zeros; the equation $u^{m}=p(t)$ defines a superelliptic curve $C$. Its coordinate ring is

the ring of regular functions on the affine part of $C$ with the fibres over $t=0$ and $t=\infty$ removed. The projection $t\colon C\to\mathbb{P}^{1}$ exhibits $C$ as an $m$-sheeted covering of the projective line, ramified over the zeros of $p$ and possibly over $t=\infty$; on its affine part $C$ is non-singular, since $p$ has no multiple zeros. By the Riemann–Hurwitz formula the genus of $C$ is

Throughout this paper we take $p$ to be the palindromic polynomial

of degree $d=2r$, which has no multiple zeros for $a\neq\pm 1$. For $m=2$ the genus is then $g=r-1$: the case $r=1$ (degree two) is rational, $r=2$ (a quartic) has genus one, and $r\geq 3$ is hyperelliptic in the usual sense. Our main results concern the case $m=2$, $r=1$.

This case merits closer description. Here $p(t)=1-2at+t^{2}$ has degree two, so the curve $u^{2}=p(t)$ is rational, and the covering $t\colon C\to\mathbb{P}^{1}$ is unramified over both $t=0$ and $t=\infty$; two points of $C$ lie over each. The algebra $A_{2}$ is therefore the ring of functions regular away from these four marked points, and $\mathfrak{sl}_{2}\otimes A_{2}$ is the four-point affine Lie algebra of Bremner [14, Prop. 1.1] (see also Cox [15]). It is not a two-point Krichever–Novikov algebra. Its universal central extension has a three-dimensional centre [14, 13]. The palindromic symmetry $t\mapsto 1/t$ of $p$ distinguishes one splitting of the four points into two pairs, hence one line in this space of local cocycles; the cocycle used throughout this paper is this symmetry-selected one. We work in the multipoint framework of Schlichenmaier [11, 12].

Given $\mathfrak{g}=\mathfrak{sl}_{2}$ (though the construction extends to any simple Lie algebra), consider the current algebra $\mathfrak{g}\otimes A_{m}$ and its universal central extension (UCE) $\widetilde{\mathfrak{g}\otimes A_{m}}$. By Kassel–Loday [2] the defining cocycle

takes its values in $\Omega^{1}_{A_{m}}/dA_{m}$, which is the centre described above and which, in the genus-zero four-point case, is three-dimensional.

Two symbols are needed here and we keep them apart, because they denote different objects. Let $c\in\Omega^{1}_{A_{m}}/dA_{m}$ span the line singled out by the palindromic symmetry, and let $c^{*}$ be the corresponding coordinate functional. For the Heisenberg-type generators introduced below we set

Thus $\gamma$ is the cocycle of the universal central extension, with values in a three-dimensional space, whereas $\psi^{(ij)}_{kl}(a)$ is a scalar: the component of $\gamma$ along one of the three available directions, namely the one the symmetry selects.

Being a component of $\gamma$, the scalar $\psi^{(ij)}_{kl}(a)$ is skew-symmetric,

for all $k,l$ and all $i,j$. This is inherited from Kassel–Loday and not an extra hypothesis. The Killing form $(\,\cdot,\cdot\,)$ is symmetric, while $\overline{f\,dg}=-\,\overline{g\,df}$ in $\Omega^{1}_{A_{m}}/dA_{m}$ because $d(fg)=f\,dg+g\,df$ is exact. Closed expressions for $\psi$ in the superelliptic setting are computed in [3]; for the case treated in this paper they are given in (3) below.

We write ${\mathcal{H}_{m}}\subset\widetilde{\mathfrak{g}\otimes A_{m}}$ for the Heisenberg subalgebra, the subalgebra spanned by the Heisenberg-type generators $\{b_{n}^{(j)}:n\in\mathbb{Z},1\leq j\leq\lfloor m/2\rfloor\}$ arising from the odd powers of $u$ in the Laurent expansion. (For $\mathfrak{g}=\mathfrak{sl}_{2}$, the relevant Chevalley generators pair to form such degrees-of-freedom.) Restricting $\gamma$ to this subalgebra gives it a central extension of its own, which we now record.

We fix once and for all that the mode index is $k$ and that the letter $m$ is reserved for the covering degree. Skew-symmetry of the brackets is not an assumption here but a consequence of (2), which is why no antisymmetry hypothesis appears in the definition. Only one case is used below, and we record it now.

For $m=2$, $r=1$ there is a single family of generators, and we drop the superscript $j=1$. Here $\psi_{kl}(a)=k\,\delta_{k+l,0}\,\omega_{1}$ with $\omega_{1}>0$ a constant depending only on $p$, so that

and all other brackets among the $b$’s vanish. In particular $b_{0}$ is central in ${\mathcal{H}_{m}}$, every bracket $[b_{k},b_{l}]$ with $k+l\neq 0$ vanishes, and ${\mathcal{H}_{m}}$ is the classical infinite-dimensional Heisenberg algebra, the constant $\omega_{1}$ amounting to a rescaling of the central element.

The triangular decomposition of ${\mathcal{H}_{m}}$ is defined by the natural grading on generators with respect to the Laurent degree $n$:

where

• •

  ${\mathcal{H}_{m}}_{+}=\bigoplus_{n>0}\mathbb{C}b_{n}^{(j)}$ (positive-degree generators),

• •

  ${\mathcal{H}_{m}}_{0}=\bigoplus_{j}\mathbb{C}b_{0}^{(j)}\oplus\mathbb{C}c$ (degree-zero Cartan),

• •

  ${\mathcal{H}_{m}}_{-}=\bigoplus_{n<0}\mathbb{C}b_{n}^{(j)}$ (negative-degree generators).

This is the standard algebraic triangular decomposition, grading by the Laurent index $n$. Geometrically, this coincides with the almost-graded structure of Schlichenmaier [9] restricted to the Heisenberg subalgebra of the universal central extension. (The alternative geometric decomposition, the in/out decomposition from the branch points, is distinct; for this paper, we use the algebraic grading.)

The triangular decomposition makes ${\mathcal{H}_{m}}$ suitable for the standard construction of Verma modules and highest-weight representations.

We work with the standard category of highest-weight ${\mathcal{H}_{m}}$-modules. An ${\mathcal{H}_{m}}$-module $M$ is called a weight module if it decomposes as

where $M_{\mu}=\{v\in M:b_{0}^{(j)}v=\mu(b_{0}^{(j)})v\text{ and }cv=\mu(c)v\}$.

A highest-weight module for ${\mathcal{H}_{m}}$ is specified by:

1. (1)

   A linear functional $\varphi:{\mathcal{H}_{m}}_{0}\to\mathbb{C}$ (the weight). No condition is imposed on $\varphi(c)$ here. The value $\varphi(c)=0$ is permitted, and Theorem 3.8 identifies it as exactly the degenerate case.

2. (2)

   A highest-weight vector $v_{\varphi}$ satisfying:

   • •

     $b_{n}^{(j)}v_{\varphi}=0$ for all $n>0$ and all $j$,

   • •

     $b_{0}^{(j)}v_{\varphi}=\varphi(b_{0}^{(j)})v_{\varphi}$,

   • •

     $cv_{\varphi}=\varphi(c)v_{\varphi}$.

3. (3)

   The action of ${\mathcal{H}_{m}}_{-}$ is locally nilpotent, i.e., for each $v\in M$ and each $n<0$, the operators $b_{n}^{(j)}$ act nilpotently on the finite-dimensional weight spaces.

Such modules are called $\varphi$-modules in the language of [17, 20], where analogous constructions for map algebras are developed.

The main object of study in this paper is the Verma module, the universal highest-weight module generated by a highest-weight vector of weight $\varphi$.

Following Definition 2.1 the bracket $[b_{k}^{(i)},b_{l}^{(j)}]$ vanishes whenever $k,l\geq 0$, since $\psi^{(ij)}_{kl}$ is supported on $k+l=0$ and the structure constant carries the factor $k$, so ${\mathcal{H}_{m}}_{+}\oplus{\mathcal{H}_{m}}_{0}$ is abelian and every linear functional on it is a character. In particular $M(\varphi)$ is defined for every $\varphi$, including $\varphi(c)=0$.

A word on terminology, since the letter $\varphi$ is used differently in part of the literature. Bekkert, Benkart, Futorny and Kashuba [6] attach a module to a polarisation, a sign function $\phi\colon\mathbb{N}\to\{\pm\}$ that decides which imaginary modes are placed in the positive part of the triangular decomposition, and call the result a $\phi$-Verma module. Here $\varphi$ is not a polarisation but a weight, a linear functional on ${\mathcal{H}_{m}}_{0}$, and the polarisation is the standard one, all positive modes annihilating $v_{\varphi}$. In the notation of [6] this is the case $\phi\equiv+$, for which their module is, in their words, the usual Verma module for the Heisenberg algebra [6, §3.1].

By the PBW theorem, $M(\varphi)$ has a $\mathbb{C}$-basis consisting of the monomials $b_{-\lambda_{1}}^{(j_{1})}\cdots b_{-\lambda_{s}}^{(j_{s})}v_{\varphi}$ with $\lambda_{1}\geq\cdots\geq\lambda_{s}>0$. In the genus-zero case there is a single family of generators, the superscripts are absent, and such a monomial is determined by the partition $\lambda=(\lambda_{1},\dots,\lambda_{s})$ alone; we then write

The grading is inherited from the degree: the weight-$(-n)$ subspace $M(\varphi)[-n]$ has basis $\{\,b_{-\lambda}v_{\varphi}:\lambda\vdash n\,\}$, of dimension the number $p(n)$ of partitions of $n$.

Denote by $G_{n}$ the Gram matrix of $\mathcal{S}_{\varphi}$ on $M(\varphi)[-n]$ in the basis (4), that is, the $p(n)\times p(n)$ matrix with entries $\mathcal{S}_{\varphi}(b_{-\lambda}v_{\varphi},\,b_{-\mu}v_{\varphi})$ for $\lambda,\mu\vdash n$. The Kac determinant analogue $\det(G_{n})$ is evaluated in closed form in Theorem 3.8, equation (10). It also gives the condition on the weight under which the constructions below behave well.

This is a condition on the determinant of an explicit matrix, so it can be checked by computation. For $m=2$, $r=1$ it collapses to the single requirement $\varphi(c)\neq 0$ (Theorem 3.8).

The bridge to orthogonal polynomials needs nothing beyond the defining polynomial of the curve and the classical uniqueness of orthogonal polynomial sequences [5]. The following lemma derives what is required for Theorem 3.7.

For $\deg p=4$ the analogous computation yields associated ultraspherical and Gegenbauer families instead, and those are the subject of [3].

We can now define the polynomial vectors precisely.

Equivalently, $\tilde{P}_{n}$ is the unique direction in $M(\varphi)[-n]$ whose Shapovalov pairing with a monomial does not depend on which monomial is taken, hence the unique direction orthogonal to all differences $b_{-\lambda}v_{\varphi}-b_{-\mu}v_{\varphi}$ with $\lambda,\mu\vdash n$.

Formula (7) is the image of the complete homogeneous symmetric function $h_{n}$ under $p_{k}\mapsto q^{-1}b_{-k}$, in the standard expansion $h_{n}=\sum_{\lambda\vdash n}p_{\lambda}/z_{\lambda}$ [4, Ch. I, (2.14)]; the weight $q^{-\ell(\lambda)}$ is what the substitution contributes. It is worth separating the parts of the theorem. Part (ii) uses only the grading. Part (iv) fixes a normalisation and needs the reality hypothesis $q\in\mathbb{R}_{>0}$, which is not automatic: $\varphi$ takes values in $\mathbb{C}$, and for $q\notin\mathbb{R}_{>0}$ the scalars $\kappa_{n}^{\prime}$ are not real and the form is not positive definite on the real span of the monomials. What is specific to ${\mathcal{H}_{m}}$ is the value $z_{\lambda}\,q^{\,\ell(\lambda)}$ in (9), produced by the cocycle of Definition 2.1. For $m\geq 3$ the cocycle acquires off-diagonal components, the Gram matrix is no longer diagonal, and neither (7) nor the evaluation in (iii) survives.

Theorem 3.8 says that the genus-zero, four-point Heisenberg algebra exhibits no new behaviour, and identifies the source. By Definition 2.1 the cocycle is supported on $k+l=0$, so the contravariant form is diagonal and its determinant cannot acquire zeros away from $\varphi(c)=0$. The interest of the Shapovalov-regularity condition of Definition 3.4 therefore lies entirely in the range $m\geq 3$, where off-diagonal cocycle components destroy the diagonality used above.

For two Heisenberg generators $b_{k},b_{l}$ ($k,l\in\mathbb{Z}$), define the normal-ordered product

i.e., negative-mode (and zero-mode) operators are placed to the left of positive-mode operators. Because $[b_{k},b_{l}]=\psi_{kl}\,c$ is a scalar on $M(\varphi)$, the normal-ordered product differs from the bare product by a scalar correction whenever $k+l=0$, and equals the bare product otherwise. In particular, the sum $\sum_{k\geq 1}{:}b_{-k}\,b_{k}{:}=\sum_{k\geq 1}b_{-k}\,b_{k}$ is already normal-ordered as written (each $-k$ is negative, so each factor stays in place).

The element ${\mathsf{L}}$ is the zero mode of the standard Sugawara stress-energy tensor of the free boson (central charge $c_{\mathrm{Vir}}=1$), the element that generates the energy grading on the free-boson Fock space; see [18, 19] for the general construction. Its eigenvalue on a level-$n$ monomial is $q\,n$, so it records the level only up to the factor $q$, which depends on the cocycle normalisation and on the central charge. Dividing by $q$ is what makes the eigenvalue the integer $n$ itself, and thereby makes the intertwining with the classical Legendre operator exact, with no parameter-dependent constants. The price is that $L_{0}$, and with it $\Omega$, are attached to $M(\varphi)$ rather than to the algebra; ${\mathsf{L}}$ is the part of the construction that is canonical.