nLab Borcherds algebra

Redirected from "Borcherds' algebras".

Context

Lie theory

∞-Lie theory (higher geometry)

Background

Smooth structure

Higher groupoids

Lie theory

∞-Lie groupoids

∞-Lie algebroids

Formal Lie groupoids

Cohomology

Homotopy

Related topics

Examples

∞\infty-Lie groupoids

∞\infty-Lie groups

∞\infty-Lie algebroids

∞\infty-Lie algebras

Contents

Idea

The concept of Borcherds algebras is a generalization of that of Kac-Moody algebras (hence also called generalized Kac-Moody algebras or Borcherds-Kac-Moody algebras, “BKM algebras”) obtained by allowing imaginary simple roots.

Recall that a (symmetrizable) Kac-Moody algebra is presented by generators and relations as being glued from copies of 𝔰𝔩 ( 2 ) \mathfrak{sl}(2) , one triple (e i,h i,f i)(e_i, h_i, f_i) for each simple root α i\alpha_i, according to a Cartan matrix (a ij)(a_{i j}) whose diagonal entries are all a ii=2a_{i i} = 2 and whose off-diagonal entries are non-positive integers, subject to the Serre relations. That all diagonal entries are positive means that all simple roots have positive norm (“real roots”). This makes the adjoint action of each of these 𝔰𝔩 ( 2 ) \mathfrak{sl}(2) -triples locally finite (“integrable”) and yields a Weyl group generated by the reflections in all the simple roots.

The idea of Borcherds 1988 is to keep this kind of presentation but to admit also diagonal entries a ii≤0a_{i i} \leq 0, hence simple roots α i\alpha_i of non-positive norm (“imaginary simple roots”). For these:

  • the triple (e i,h i,f i)(e_i, h_i, f_i) still spans a copy of 𝔰𝔩 ( 2 ) \mathfrak{sl}(2) if a ii<0a_{i i} \lt 0, but a copy of the 3-dimensional Heisenberg Lie algebra if a ii=0a_{i i} = 0;

  • there are no Serre relations of the form (ad e i) n(e j)=0(\mathrm{ad}_{e_i})^n(e_j) = 0 (so that e ie_i need not act locally nilpotently), only the relation [e i,e j]=0=[f i,f j][e_i, e_j] = 0 = [f_i, f_j] whenever a ij=0a_{i j} = 0;

  • there is no corresponding reflection: the Weyl group is generated by the reflections in the real simple roots only.

The effect is that Borcherds algebras behave like Kac-Moody algebras with some “freely generated directions” adjoined:

For example, if no two imaginary simple roots are orthogonal, then a Borcherds algebra decomposes as a direct sum 𝔲 −⊕(𝔤 J+𝔥)⊕𝔲 +\mathfrak{u}^- \oplus (\mathfrak{g}_J + \mathfrak{h}) \oplus \mathfrak{u}^+, where 𝔤 J\mathfrak{g}_J is the Kac-Moody algebra of the real simple roots, 𝔥\mathfrak{h} is the Cartan subalgebra, and 𝔲 ±\mathfrak{u}^\pm are free Lie algebras (Jurisich 1998).

Nevertheless, most of the structure theory of Kac-Moody algebras carries over (Borcherds 1988, Kac 1990, §11.13), notably the invariant bilinear form, the root space decomposition with finite-dimensional root spaces, the Weyl group, and a version of the Weyl-Kac character formula and its denominator identity, now with an extra correction term contributed by the imaginary simple roots.

More conceptually, Borcherds algebras are essentially characterized as the Lie algebras carrying an almost positive definite contravariant form (Borcherds 1988, §3, Borcherds 1995a): If 𝔤=⨁ n∈ℤ𝔤 n\mathfrak{g} = \bigoplus_{n \in \mathbb{Z}} \mathfrak{g}_n is a ℤ\mathbb{Z}-graded Lie algebra with 𝔤 n\mathfrak{g}_n finite-dimensional for n≠0n \neq 0, equipped with an involution ω\omega which maps 𝔤 n\mathfrak{g}_n to 𝔤 −n\mathfrak{g}_{-n} and acts as −1-1 on 𝔤 0\mathfrak{g}_0, and with an ω\omega-invariant symmetric invariant bilinear form (−,−)(-,-) which pairs 𝔤 n\mathfrak{g}_n only with 𝔤 −n\mathfrak{g}_{-n}, such that the contravariant form

(x,y) 0≔−(x,ω(y)) (x,y)_0 \;\coloneqq\; -\big(x, \omega(y)\big)

is positive definite on all 𝔤 n≠0\mathfrak{g}_{n \neq 0}, then the kernel of (−,−)(-,-) is central and the quotient by it is a Borcherds algebra (Borcherds 1988, Thm. 3.1).

Conversely, Borcherds algebras satisfy these conditions, up to minor technical caveats (Borcherds 1995a, §1). This is just the structure carried by spaces of physical states in string theory, where positivity of the contravariant form is the content of the no-ghost theorem (Borcherds 1988, §5). This is how Borcherds algebras were found (Borcherds 1986) and why they keep appearing in string theory.

Definition

We follow Borcherds 1995a, §1. All Lie algebras are over the real numbers.

Definition

(Borcherds-Cartan matrix)

A Borcherds-Cartan matrix is a real square matrix A=(a ij) i,j∈IA = (a_{i j})_{i,j \in I}, indexed by a countable set II, such that:

  1. a ij=a jia_{i j} = a_{j i} for all i,j∈Ii, j \in I;

  2. a ij≤0a_{i j} \leq 0 for all i≠ji \neq j;

  3. if a ii>0a_{i i} \gt 0 then 2a ij/a ii∈ℤ2 a_{i j}/a_{i i} \in \mathbb{Z} for all j∈Ij \in I.

An index i∈Ii \in I is called real if a ii>0a_{i i} \gt 0, and imaginary if a ii≤0a_{i i} \leq 0 (isotropic if moreover a ii=0a_{i i} = 0).

Definition

(universal Borcherds algebra)

Given a Borcherds-Cartan matrix AA (Def. ), the universal Borcherds algebra 𝔤^(A)\widehat{\mathfrak{g}}(A) is the Lie algebra generated by elements

e i,f i,h ij,i,j∈I, e_i,\, f_i,\, h_{i j} \,, \;\;\;\; i, j \in I \,,

subject to the following relations, for all i,j,k∈Ii, j, k \in I:

  1. [e i,f j]=h ij[e_i, f_j] = h_{i j};

  2. [h ij,e k]=δ ija ike k[h_{i j}, e_k] = \delta_{i j}\, a_{i k}\, e_k and [h ij,f k]=−δ ija ikf k[h_{i j}, f_k] = - \delta_{i j}\, a_{i k}\, f_k;

  3. if a ii>0a_{i i} \gt 0 and j≠ij \neq i, then (ad e i) 1−2a ij/a ii(e j)=0(\mathrm{ad}_{e_i})^{1 - 2 a_{i j}/a_{i i}}(e_j) = 0 and (ad f i) 1−2a ij/a ii(f j)=0(\mathrm{ad}_{f_i})^{1 - 2 a_{i j}/a_{i i}}(f_j) = 0;

  4. if a ij=0a_{i j} = 0, then [e i,e j]=0[e_i, e_j] = 0 and [f i,f j]=0[f_i, f_j] = 0.

Remark

(the elements h ijh_{i j})

The relations of Def. imply that the h ijh_{i j} all commute with each other, that the h ijh_{i j} with i≠ji \neq j are central, and that h ij=0h_{i j} = 0 unless the iith and the jjth column of AA coincide (Borcherds 1991, see Gannon). Hence, if no two columns of AA coincide, then 𝔤^(A)\widehat{\mathfrak{g}}(A) is generated by the e ie_i, f if_i and h i≔h iih_i \coloneqq h_{i i} subject to

[h i,h j]=0,[e i,f j]=δ ijh i,[h i,e j]=a ije j,[h i,f j]=−a ijf j [h_i, h_j] = 0 \,, \;\;\; [e_i, f_j] = \delta_{i j}\, h_i \,, \;\;\; [h_i, e_j] = a_{i j}\, e_j \,, \;\;\; [h_i, f_j] = - a_{i j}\, f_j

together with items 3 and 4 of Def. . This is of the familiar form of the Chevalley-Serre presentation of Kac-Moody algebras. In general, the quotient of 𝔤^(A)\widehat{\mathfrak{g}}(A) by the h ijh_{i j} with i≠ji \neq j is what is often taken as the definition of the Borcherds algebra 𝔤(A)\mathfrak{g}(A) (e.g. Gannon). The h ijh_{i j} with i≠ji \neq j matter only for central extensions (Borcherds 1991, Borcherds 1995a, §1).

Definition

(Borcherds algebra)

A Borcherds algebra (or generalized Kac-Moody algebra) is a Lie algebra of the form

𝔤=𝔞⋊𝔡 \mathfrak{g} \;=\; \mathfrak{a} \rtimes \mathfrak{d}

where

  • 𝔞\mathfrak{a} is the quotient of a universal Borcherds algebra 𝔤^(A)\widehat{\mathfrak{g}}(A) (Def. ) by a subspace of its center,

  • 𝔡\mathfrak{d} is an abelian Lie algebra acting on 𝔞\mathfrak{a} by derivations such that all the e ie_i and f if_i are eigenvectors.

In other words, Borcherds algebras are obtained from universal Borcherds algebras by discarding part of the center and adjoining commuting outer derivations.

Remark

(roots, Weyl group and contravariant form)

Let Q≔⨁ i∈Iℤα iQ \coloneqq \bigoplus_{i \in I} \mathbb{Z}\, \alpha_i be the free abelian group on symbols α i\alpha_i (the simple roots), equipped with the bilinear form given by (α i,α j)≔a ij(\alpha_i, \alpha_j) \coloneqq a_{i j}. Declaring e ie_i to be of degree α i\alpha_i, f if_i of degree −α i-\alpha_i and h ijh_{i j} of degree 00 makes (universal) Borcherds algebras QQ-graded. A root is a nonzero α∈Q\alpha \in Q whose degree-α\alpha subspace is non-trivial. Every root is a non-negative or a non-positive integer combination of simple roots, and a root is called real if (α,α)>0(\alpha, \alpha) \gt 0 and imaginary otherwise (Borcherds 1988, §1). The root spaces are finite-dimensional, and the simple root spaces are 1-dimensional. Moreover:

  • for each i∈Ii \in I the elements e ie_i, h ih_i, f if_i span a copy of 𝔰𝔩(2,ℝ)\mathfrak{sl}(2,\mathbb{R}) if a ii≠0a_{i i} \neq 0, and a copy of the 3-dimensional Heisenberg Lie algebra if a ii=0a_{i i} = 0;

  • the Weyl group is the group of automorphisms of QQ generated by the reflections α↦α−2(α,α i)a iiα i\alpha \mapsto \alpha - \frac{2 (\alpha, \alpha_i)}{a_{i i}} \alpha_i in the real simple roots α i\alpha_i only;

  • the Cartan involution ω\omega is given by ω(e i)=−f i\omega(e_i) = - f_i and ω(f i)=−e i\omega(f_i) = - e_i, hence ω(h i)=−h i\omega(h_i) = - h_i.

For the Borcherds algebras as in Borcherds 1988, §1 (with Cartan subalgebra a real vector space HH containing the h ih_i, equipped with a symmetric bilinear form with (h i,h j)=a ij(h_i, h_j) = a_{i j}), there is a unique invariant symmetric bilinear form extending that on HH. The corresponding contravariant form (x,y) 0≔−(x,ω(y))(x,y)_0 \coloneqq -(x, \omega(y)) is positive definite on all root spaces (Borcherds 1988, Cor. 2.5, whose proof is completed in Kac 1990, §11.13, see Borcherds 1995a, §3). This is the “almost positivity” from the Idea-section.

Remark

(variants of the definition)

  • (Kac-Moody algebras as special cases) If a ii>0a_{i i} \gt 0 for all i∈Ii \in I, then item 4 in Def. follows from item 3 (Borcherds 1988, §1). After rescaling to a ii=2a_{i i} = 2 (as below), the Borcherds algebras of Def. are then the central quotients, extended by derivations, of symmetrizable Kac-Moody algebras.

  • (normalization) Rescaling the iith row of a Borcherds-Cartan matrix by 2/a ii2/a_{i i} for all real ii turns it into a symmetrizable matrix with a ii=2a_{i i} = 2 for real ii, and with a ij∈ℤa_{i j} \in \mathbb{Z} whenever a ii=2a_{i i} = 2. This is the normalization convention used, e.g., in Gannon. Kac 1990, §11.13 moreover admits non-symmetrizable matrices, for which however there is no invariant bilinear form (Borcherds 1995a, §1). For symmetrizable matrices, Kac’s definition agrees essentially with Def. (by Kac 1990, Thm. 11.13.1, see Borcherds 1995a, §3).

  • (superalgebras) There is an evident generalization to Lie superalgebras, where some of the generators e ie_i, f if_i are odd. These Borcherds superalgebras (or generalized Kac-Moody superalgebras) were envisaged already in Borcherds 1988, §5 and developed in Ray 1995, Ray 2006. In the super case, item 4 of Def. with i=ji = j has content for odd isotropic ii, where it says that [e i,e i]=0[e_i, e_i] = 0 (see Prop. for an example). It is mostly Borcherds superalgebras which appear in supergravity (see below).

Examples

Lie algebras of physical string states

The archetypical examples of Borcherds algebras, which also motivated the concept, are Lie algebras of physical quantum states of the bosonic string (Frenkel 1985, Borcherds 1986, 1988, §5). They are constructed from vertex algebras as follows (for exposition see Gebert 1993, §3.4):

Definition

(Lie algebra of physical states)

Let VV be a vertex algebra with conformal vector, whose Virasoro modes we denote by L nL_n, and with an invariant symmetric bilinear form. Its spaces of physical states of weight kk are

P k≔{v∈V|L 0v=kv,L nv=0for alln>0}. P^k \;\coloneqq\; \big\{ v \in V \,\big\vert\, L_0 v = k\, v ,\; L_n v = 0 \;\text{for all}\; n \gt 0 \big\} \,.

Then L −1P 0⊂P 1L_{-1} P^0 \subset P^1, and the quotient P 1/L −1P 0P^1 / L_{-1} P^0 is a Lie algebra with bracket given by the zero-modes of the vertex operators, [u,v]≔u 0v[u, v] \coloneqq u_0 v (Borcherds 1986, see Borcherds 1992, §6). The bilinear form descends to an invariant form on this Lie algebra, and we write

𝔤(V)≔(P 1/L −1P 0)/ker(−,−) \mathfrak{g}(V) \;\coloneqq\; \big( P^1 / L_{-1} P^0 \big) \big/ \mathrm{ker}(-,-)

for its quotient by the kernel of this form (the “null states”).

In string-theoretic terms, P 1P^1 consists of the on-shell states of the chiral string, and the bracket [u,v]=u 0v[u,v] = u_0 v is the action of the charge u 0=∮Y(u,z)dzu_0 = \oint Y(u,z)\, d z of the weight-1 current corresponding to uu.

For VV of central charge 26, the critical value for the bosonic string, the no-ghost theorem (Goddard & Thorn 1972) identifies the physical states with “transversal” states. Concretely (Borcherds 1992, Thm. 5.1), suppose that WW is a vertex operator algebra of central charge 24 with positive definite invariant bilinear form, and that

V=W⊗V II 1,1 V \;=\; W \otimes V_{\mathrm{II}_{1,1}}

is its tensor product with the lattice vertex algebra of the even unimodular Lorentzian lattice II 1,1\mathrm{II}_{1,1} of rank 2. Then 𝔤(V)\mathfrak{g}(V) is graded by II 1,1\mathrm{II}_{1,1}, with root spaces

𝔤(V) r≃W 1−r 2/2,0≠r∈II 1,1, \mathfrak{g}(V)_r \;\simeq\; W_{1 - r^2/2} \,, \;\;\;\; 0 \neq r \in \mathrm{II}_{1,1} \,,

the subspaces of WW of L 0L_0-eigenvalue 1−r 2/21 - r^2/2. This identification is equivariant for any group of automorphisms of WW. In particular, the contravariant form on 𝔤(V)\mathfrak{g}(V) is then positive definite on all nonzero root spaces, so that 𝔤(V)\mathfrak{g}(V) is a Borcherds algebra by the characterization recalled above. More generally, for every even Lorentzian lattice LL, the Lie algebra 𝔤(V L)\mathfrak{g}(V_L) of physical states of the lattice vertex algebra V LV_L is a Borcherds algebra (Borcherds 1988, §5, Borcherds 1995a, Ex. 2).

The two most prominent examples are the following.

Example

(fake Monster Lie algebra)

The fake Monster Lie algebra is

𝔤(V II 25,1), \mathfrak{g}\big( V_{\mathrm{II}_{25,1}} \big) \,,

the Lie algebra of physical states of the lattice vertex algebra of the even unimodular Lorentzian lattice II 25,1\mathrm{II}_{25,1} of rank 26 (Borcherds 1990, where it is still called “the monster Lie algebra”). Hence it is the Lie algebra of physical states of the chiral bosonic string on the 26-dimensional Lorentzian torus ℝ 25,1/II 25,1\mathbb{R}^{25,1}/\mathrm{II}_{25,1}.

Its Cartan subalgebra is II 25,1⊗ℝ\mathrm{II}_{25,1} \otimes \mathbb{R}, and by the no-ghost theorem the multiplicity of any nonzero r∈II 25,1r \in \mathrm{II}_{25,1} as a root is p 24(1−r 2/2)p_{24}(1 - r^2/2), where

∑ np 24(n)q n=∏ k≥1(1−q k) −24. \sum_{n} p_{24}(n)\, q^n \;=\; \prod_{k \geq 1} (1 - q^k)^{-24} \,.

So the roots are the nonzero rr with r 2≤2r^2 \leq 2. Its simple roots are (Borcherds 1990, Thm. 1):

  • the real simple roots: the norm-2 vectors r∈II 25,1r \in \mathrm{II}_{25,1} with (r,ρ)=−1(r, \rho) = -1, for ρ\rho a norm-0 Weyl vector; these correspond to the points of the Leech lattice (Conway 1983);

  • the imaginary simple roots: the positive multiples nρn \rho, n≥1n \geq 1, each with multiplicity 24.

Accordingly, its denominator identity reads

e ρ∏ r>0(1−e r) p 24(1−r 2/2)=∑ w∈Wdet(w)w(e ρ∏ n≥1(1−e nρ) 24), e^\rho \prod_{r \gt 0} \big(1 - e^r\big)^{p_{24}(1 - r^2/2)} \;=\; \sum_{w \in W} \det(w)\, w\Big( e^\rho \prod_{n \geq 1} \big(1 - e^{n \rho}\big)^{24} \Big) \,,

where the product is over the positive roots (the positive multiples of ρ\rho and the roots rr with (r,ρ)<0(r, \rho) \lt 0), and where WW is the Weyl group, here the reflection group of II 25,1\mathrm{II}_{25,1} generated by the reflections in its norm-2 vectors. On the right, e ρ∏ n≥1(1−e nρ) 24=∑ nτ(n)e nρe^\rho \prod_{n \geq 1}(1 - e^{n\rho})^{24} = \sum_{n} \tau(n)\, e^{n \rho} is the modular discriminant? in the variable e ρe^\rho, with τ\tau the Ramanujan tau function.

Example

(Monster Lie algebra)

The Monster Lie algebra is (Borcherds 1992, §6)

𝔪≔𝔤(V ♮⊗V II 1,1), \mathfrak{m} \;\coloneqq\; \mathfrak{g}\big( V^\natural \otimes V_{\mathrm{II}_{1,1}} \big) \,,

the Lie algebra of physical states for the moonshine module V ♮V^\natural (Frenkel, Lepowsky & Meurman 1988) tensored with the lattice vertex algebra of II 1,1≃ℤ 2\mathrm{II}_{1,1} \simeq \mathbb{Z}^2, with inner product ((m,n),(m′,n′))=−(mn′+m′n)\big((m,n),(m',n')\big) = -(m n' + m' n). Hence it is the Lie algebra of physical states of the chiral bosonic string on the product of the “Monster CFT” with a 2-dimensional Lorentzian torus.

Here V ♮=⨁ k≥0V k ♮V^\natural = \bigoplus_{k \geq 0} V^\natural_k is a vertex operator algebra of central charge 24 with positive definite invariant form, whose automorphism group is the monster group 𝕄\mathbb{M}, and whose graded dimension is the normalized j-invariant:

∑ k≥0dim(V k ♮)q k−1=j(q)−744=∑ n≥−1c(n)q n=q −1+196884q+21493760q 2+⋯. \sum_{k \geq 0} \mathrm{dim}\big( V^\natural_k \big)\, q^{k-1} \;=\; j(q) - 744 \;=\; \sum_{n \geq -1} c(n)\, q^n \;=\; q^{-1} + 196884\, q + 21493760\, q^2 + \cdots \,.

Therefore, by the no-ghost theorem (as above), 𝔪\mathfrak{m} is ℤ 2\mathbb{Z}^2-graded with root spaces

𝔪 (m,n)≃V 1+mn ♮for(m,n)≠(0,0) \mathfrak{m}_{(m,n)} \;\simeq\; V^\natural_{1 + m n} \;\;\; \text{for} \; (m,n) \neq (0,0)

as 𝕄\mathbb{M}-representations, hence of dimension c(mn)c(m n), while the Cartan subalgebra 𝔪 (0,0)\mathfrak{m}_{(0,0)} is 2-dimensional (Borcherds 1992, Thm. 6.1). The unique real simple root is (1,−1)(1,-1), of norm 2, and the imaginary simple roots are the (1,n)(1,n) for n≥1n \geq 1, of norm −2n-2n and with multiplicity c(n)c(n) (Borcherds 1992, Thm. 7.2). Accordingly, writing pp and qq for the formal exponentials of the degrees (1,0)(1,0) and (0,1)(0,1), the denominator identity of 𝔪\mathfrak{m} is the product formula

p −1∏ m>0,n∈ℤ(1−p mq n) c(mn)=j(p)−j(q). p^{-1} \prod_{m \gt 0, n \in \mathbb{Z}} \big(1 - p^m q^n\big)^{c(m n)} \;=\; j(p) - j(q) \,.

Since the Monster group acts on 𝔪\mathfrak{m} with root spaces the 𝕄\mathbb{M}-representations V 1+mn ♮V^\natural_{1 + m n}, there is for each g∈𝕄g \in \mathbb{M} a twisted denominator identity (Borcherds 1992, (8.3)):

p −1exp(−∑ i≥1∑ m≥1,n∈ℤTr(g i|V 1+mn ♮)p miq nii)=T g(p)−T g(q), p^{-1} \exp\Big( - \sum_{i \geq 1} \sum_{m \geq 1, n \in \mathbb{Z}} \mathrm{Tr}\big(g^i \,\vert\, V^\natural_{1 + m n}\big)\, \frac{p^{m i}\, q^{n i}}{i} \Big) \;=\; T_g(p) - T_g(q) \,,

where

T g(q)≔∑ n≥−1Tr(g|V n+1 ♮)q n T_g(q) \;\coloneqq\; \sum_{n \geq -1} \mathrm{Tr}\big(g \,\vert\, V^\natural_{n+1}\big)\, q^n

is the McKay-Thompson series of gg. For g=eg = e this reduces to the previous identity, since T e=j−744T_e = j - 744. These identities imply recursion relations (“replication formulas”) which determine each T gT_g from its first few coefficients. Checking these coefficients proves that the T gT_g are the Hauptmoduln predicted by Conway and Norton, hence the monstrous moonshine conjecture (Borcherds 1992, Thm. 1.1). For the relation of replicability to Hecke operators and power operations in equivariant elliptic cohomology and Tate K-theory see Ganter 2009, 2013.

In both examples, the denominator identity is the product expansion of an automorphic form. For the fake Monster Lie algebra this is Borcherds’ automorphic form Φ 12\Phi_{12} of weight 12 for the orthogonal group of the lattice II 26,2\mathrm{II}_{26,2} (Borcherds 1995b). For the Monster Lie algebra it is j(p)−j(q)j(p) - j(q), a modular function in each of the variables pp and qq. More generally, the denominator identities of Borcherds algebras are often automorphic forms on orthogonal groups O(s+2,2)O(s+2,2) with infinite product expansions (“Borcherds products”, Borcherds 1995b, 1998). Conversely, hyperbolic Kac-Moody algebras may be enlarged (“automorphically corrected”) to Borcherds (super)algebras whose denominator function is an automorphic form (Gritsenko & Nikulin 1997).

Algebras of BPS States

In string theory, Borcherds algebras appear as algebras of BPS states (Harvey & Moore 1996, 1998). For instance, the degeneracies of dyons in 𝒩=4\mathcal{N}=4 string compactifications are given in terms of the denominator of a Borcherds superalgebra (Dijkgraaf, Verlinde & Verlinde 1997), whose Weyl group governs their wall crossing (Cheng & Verlinde 2008). Closely related, the generating function of elliptic genera of symmetric products, which is the partition function of second quantized strings, is (almost) an automorphic form for O(3,2;ℤ)O(3,2;\mathbb{Z}) (Dijkgraaf, Moore, Verlinde & Verlinde 1997). The Monster Lie algebra itself appears as an algebra of BPS states in certain heterotic string models (Paquette, Persson & Volpato 2016).

In a mathematical formalization, the BPS algebras of 2-Calabi-Yau categories, such as those of semistable coherent sheaves on K3 surfaces, of semistable Higgs bundles on curves, or of modules over preprojective algebras, are the positive parts of the enveloping algebras of generalized Kac-Moody algebras (Davison, Hennecart & Schlegel Mejia 2023).

Organizing SuGra higher gauge fields

In supergravity, Borcherds superalgebras organize the p-form higher gauge fields together with their duals, via the Bianchi identities of their flux densities.

The mechanism

The Bianchi identities of the flux densities in supergravity are all of the schematic form “dF=F∧F\mathrm{d} F = F \wedge F”, and as such they encode a Lie superalgebra (Cremmer, Julia, Lü & Pope 1998a, b):

To each flux density F aF^a of degree deg(F a)=p a+1\mathrm{deg}(F^a) = p_a + 1 (the field strength of a p ap_a-form potential) associate a generator t at_a of degree p ap_a, which is even or odd according to the parity of p ap_a. Then all the Bianchi identities combine into the single equation

d𝔽=12[𝔽,𝔽],𝔽≔∑ aF at a, \mathrm{d} \mathbb{F} \;=\; \frac{1}{2} [\mathbb{F}, \mathbb{F}] \,, \;\;\;\;\; \mathbb{F} \;\coloneqq\; \textstyle{\sum_a} F^a\, t_a \,,

where we use the Koszul sign rule

[F at a,F bt b]≔(−1) |t a|deg(F b)F aF b[t a,t b] \big[ F^a \, t_a ,\, F^b \, t_b \big] \;\coloneqq\; (-1)^{\vert t_a \vert \, \mathrm{deg}(F^b)} \, F^a \, F^b \, [t_a, t_b]

and the usual (left) Leibniz rule for d\mathrm{d}. This equation defines a graded skew-symmetric bracket on the t at_a, and consistency of the Bianchi identities, d 2𝔽=0\mathrm{d}^2 \mathbb{F} = 0, is equivalent to the graded Jacobi identity for this bracket (Greitz & Howe 2011b, §1).

In Cremmer, Julia, Lü & Pope 1998b this leads to a “doubled” formulation of maximal supergravities, in which the equations of motion become a twisted self-duality condition on the total field strength, valued in a finite-dimensional Lie superalgebra. These “duality superalgebras” turn out to be truncations of infinite-dimensional Borcherds superalgebras, which may be derived from E 11 E_{11} (Henneaux, Julia & Levie 2012) and whose root lattices, for toroidal KK-compactifications of M-theory, are the second integral cohomology lattices of del Pezzo surfaces (Henry-Labordère, Julia & Paulot 2002, 2003, cf. mysterious duality).

The truncation reflects the fact that on ordinary spacetime all forms of degree beyond the spacetime dimension vanish. On superspace, where the odd basis 1-forms commute with each other, there are non-vanishing forms of arbitrary degree, and the algebra of all consistent flux densities becomes the full positive part 𝔫 +\mathfrak{n}^+ of the Borcherds superalgebra (Greitz & Howe 2011b). Moreover, a superspace cohomology argument, using the vanishing of the cohomology groups H t p,qH^{p,q}_t of the dimension-zero torsion for p>1p \gt 1 (Berkovits & Howe 2008), shows (Greitz & Howe 2011b, §2) that

  • once the Bianchi identities of the physical fields hold, those of all the higher-degree flux densities are automatically solvable,

  • there are no “gauge-trivial” flux densities (those with dF=0\mathrm{d} F = 0) apart from exact 11-forms.

Type IIA

The case of type IIA supergravity illustrates the mechanism most transparently:

Example

(form algebra of type IIA supergravity)

The bosonic flux densities of (massless) type IIA supergravity are

  • the RR-flux densities F 2F_2, F 4F_4 and their duals F 6F_6, F 8F_8;

  • the NS-flux H 3H_3 (of the B-field), its dual H 7H_7, and the 9-flux H 9H_9 dual to the dilaton;

  • the RR 10-flux F 10F_{10}, dual to the Romans mass F 0F_0, and NS 11-fluxes H 11H_{11}, the field strengths of 10-form potentials.

With wedge products understood, their Bianchi identities read (Greitz & Howe 2011b, (2.13-14)):

dH 3 = 0 dF 2k+2 = H 3F 2k,k=0,⋯,4, F 0=0 dH 7 = 12F 4F 4−F 2F 6 dH 9 = +H 3H 7+12F 4F 6−32F 2F 8, \begin{array}{rcl} \mathrm{d} H_3 &=& 0 \\ \mathrm{d} F_{2k+2} &=& H_3 \, F_{2k} \mathrlap{\,, \;\;\;\;\; \begin{array}{l} k = 0, \cdots, 4 \,, \\ F_0 = 0 \end{array} } \\ \mathrm{d} H_7 &=& \frac{1}{2} F_4 F_4 - F_2 F_6 \\ \mathrm{d} H_9 &=& + H_3 H_7 + \frac{1}{2} F_4 F_6 - \frac{3}{2} F_2 F_8 \mathrlap{\,,} \end{array}

together with a 2-parameter family of consistent Bianchi identities for H 11H_{11} (see the table in Remark below). Here the sign of the term H 3H 7H_3 H_7 differs from that in Greitz & Howe 2011b, (2.14), where d\mathrm{d} acts from the right, as is customary in superspace. In the Romans-massive case, F 0=m≠0F_0 = m \neq 0, these identities are modified to dF 2=F 0H 3\mathrm{d} F_2 = F_0 \, H_3, and in the NS identities F 2F 2nF_2 \, F_{2n} is replaced by F 2F 2n−F 0F 2n+2F_2\, F_{2n} - F_0 F_{2n+2} for n≥3n \geq 3 (Greitz & Howe 2011b, §2.2, see also Lavrinenko, Lü, Pope & Stelle 1999).

To find the corresponding Lie superalgebra (Greitz & Howe 2011b, §4.2), assign the even generator e 0e_0 to H 3H_3 and the odd generator e 1e_1 to F 2F_2. Since H 3H_3 is the only 3-flux, we have [e 1,e 1]=ce 0[e_1, e_1] = c\, e_0 for some constant cc, and the corresponding component of d𝔽=12[𝔽,𝔽]\mathrm{d} \mathbb{F} = \frac{1}{2}[\mathbb{F}, \mathbb{F}] is dH 3=c2F 2F 2\mathrm{d} H_3 = \frac{c}{2} F_2 \, F_2. Therefore dH 3=0\mathrm{d} H_3 = 0 requires that

[e 1,e 1]=0. [e_1, e_1] \;=\; 0 \,.

The generators of the higher fluxes are then iterated brackets of e 0e_0 and e 1e_1:

  • the RR-fluxes correspond to F 2k↔(ad e 0) k−1(e 1)F_{2k} \leftrightarrow (\mathrm{ad}_{e_0})^{k-1}(e_1), for all k≥1k \geq 1;

  • the further NS-fluxes arise from brackets of these, starting with H 7↔[[e 0,e 1],[e 0,e 1]]H_7 \leftrightarrow \big[[e_0, e_1], [e_0, e_1]\big].

The smallest Borcherds superalgebra (in the sense of Remark ) which accommodates these brackets has generators e i,h i,f ie_i, h_i, f_i for i∈{0,1}i \in \{0,1\}, with e 1e_1 and f 1f_1 odd, and Cartan matrix

A=(0 −1 −1 0). A \;=\; \left( \array{ 0 & -1 \\ -1 & 0 } \right) \,.

Its entries are fixed as follows:

  • a 11=0a_{1 1} = 0 is forced by [e 1,e 1]=0[e_1, e_1] = 0: For odd e 1e_1, f 1f_1 the super Jacobi identity gives

    [f 1,[e 1,e 1]]=[h 1,e 1]−[e 1,h 1]=2a 11e 1, \big[f_1, [e_1, e_1]\big] \;=\; [h_1, e_1] - [e_1, h_1] \;=\; 2\, a_{1 1}\, e_1 \mathrlap{\,,}

    so that a 11≠0a_{1 1} \neq 0 would imply [e 1,e 1]≠0[e_1, e_1] \neq 0. Conversely, for a 11=0a_{1 1} = 0 the relation [e 1,e 1]=0[e_1, e_1] = 0 is part of the definition (item 4 of Def. with i=ji = j, which has content only for odd generators);

  • a 00=0a_{0 0} = 0 is forced by the fact that (ad e 0) n(e 1)≠0(\mathrm{ad}_{e_0})^n(e_1) \neq 0 for all nn, while there are only finitely many forms of each degree: a 00>0a_{0 0} \gt 0 would contradict the first fact, and a 00<0a_{0 0} \lt 0 would produce a copy of 𝔰𝔩(2)\mathfrak{sl}(2) with infinite-dimensional representations, contradicting the second;

  • a 01=a 10=−1a_{0 1} = a_{1 0} = -1 is a normalization.

Both simple roots are isotropic, so that the triples (e i,h i,f i)(e_i, h_i, f_i) span Heisenberg (super-)algebras.

The positive part of this Borcherds superalgebra is as free as it can be:

Proposition

(positive part of the IIA Borcherds superalgebra)

Let 𝔤\mathfrak{g} be the Borcherds superalgebra of Example , and let 𝔫 +⊂𝔤\mathfrak{n}^+ \subset \mathfrak{g} be the sub-algebra generated by e 0e_0 and e 1e_1.

  1. 𝔫 +\mathfrak{n}^+ is the free Lie superalgebra 𝕃(e 0,e 1)\mathbb{L}(e_0, e_1) on an even generator e 0e_0 and an odd generator e 1e_1, modulo the super-Lie ideal generated by [e 1,e 1][e_1, e_1]:

    𝔫 +≃𝕃(e 0,e 1)/⟨[e 1,e 1]⟩. \mathfrak{n}^+ \;\simeq\; \mathbb{L}(e_0, e_1) \big/ \big\langle [e_1, e_1] \big\rangle \,.
  2. No further relations appear if 𝔤\mathfrak{g} is understood as the contragredient Lie superalgebra of AA, hence as the quotient by the maximal ideal which intersects the Cartan subalgebra trivially.

  3. Writing x k≔(ad e 0) k(e 1)x_k \coloneqq (\mathrm{ad}_{e_0})^k(e_1), for k≥1k \geq 1, there are isomorphisms

    𝔫 +≃ℝe 1⋉𝕃(e 0,x 1)≃ℝe 1⋉(ℝe 0⋉𝕃(x 1,x 2,x 3,⋯)), \mathfrak{n}^+ \;\simeq\; \mathbb{R} e_1 \ltimes \mathbb{L}\big(e_0, x_1\big) \;\simeq\; \mathbb{R} e_1 \ltimes \Big( \mathbb{R} e_0 \ltimes \mathbb{L}\big(x_1, x_2, x_3, \cdots\big) \Big) \,,

    where 𝕃(⋯)\mathbb{L}(\cdots) denotes free Lie superalgebras on the given generators, all of which except e 0e_0 are odd. In particular, as a graded vector space

    𝔫 +≃ℝe 0⊕ℝe 1⊕𝕃(x 1,x 2,x 3,⋯). \mathfrak{n}^+ \;\simeq\; \mathbb{R} e_0 \oplus \mathbb{R} e_1 \oplus \mathbb{L}\big(x_1, x_2, x_3, \cdots\big) \,.
  4. The universal enveloping algebra of 𝔫 +\mathfrak{n}^+ is the free associative algebra on e 0e_0 and e 1e_1 modulo e 1 2=0e_1^2 = 0. Its subspace of degree nα 0+mα 1n \alpha_0 + m \alpha_1 has a basis given by the words with nn letters e 0e_0 and mm non-adjacent letters e 1e_1, of which there are (n+1 m)\left(\array{n+1 \\ m}\right). Hence the root multiplicities mult(n,m)≔dim(𝔫 nα 0+mα 1 +)\mathrm{mult}(n,m) \coloneqq \mathrm{dim}\big(\mathfrak{n}^+_{n \alpha_0 + m \alpha_1}\big) are determined by

    ∏ n,m≥0(1−(−1) mp nq m) −(−1) mmult(n,m)=1+q1−p−pq. \prod_{n, m \geq 0} \Big( 1 - (-1)^m \, p^n q^m \Big)^{ - (-1)^m \, \mathrm{mult}(n,m) } \;=\; \frac{1 + q}{1 - p - p q} \,.
Proof

Item 1: Let 𝔤˜\widetilde{\mathfrak{g}} be the Lie superalgebra generated by e i,f i,h ie_i, f_i, h_i, for i∈{0,1}i \in \{0,1\} and with e 1,f 1e_1, f_1 odd, subject only to the relations

[h i,h j]=0,[e i,f j]=δ ijh i,[h i,e j]=a ije j,[h i,f j]=−a ijf j. [h_i, h_j] = 0 \,, \;\; [e_i, f_j] = \delta_{i j}\, h_i \,, \;\; [h_i, e_j] = a_{i j}\, e_j \,, \;\; [h_i, f_j] = - a_{i j}\, f_j \,.

(Since the two columns of AA differ, these are items 1 and 2 of Def. , by Remark .) As for Kac-Moody algebras (Kac 1990, Thm. 1.2), whose proof carries over verbatim to the super case, 𝔤˜\widetilde{\mathfrak{g}} has a triangular decomposition

𝔤˜=𝔫˜ −⊕𝔥⊕𝔫˜ + \widetilde{\mathfrak{g}} \;=\; \widetilde{\mathfrak{n}}^- \oplus \mathfrak{h} \oplus \widetilde{\mathfrak{n}}^+

with 𝔫˜ +≃𝕃(e 0,e 1)\widetilde{\mathfrak{n}}^+ \simeq \mathbb{L}(e_0, e_1) and 𝔫˜ −≃𝕃(f 0,f 1)\widetilde{\mathfrak{n}}^- \simeq \mathbb{L}(f_0, f_1) free. Item 3 of Def. is void, since no index is real, and item 4 amounts to [e 1,e 1]=0=[f 1,f 1][e_1, e_1] = 0 = [f_1, f_1], since [e 0,e 0]=0=[f 0,f 0][e_0, e_0] = 0 = [f_0, f_0] holds anyway, e 0e_0 and f 0f_0 being even, while a 01≠0a_{0 1} \neq 0.

Let I +⊂𝔫˜ +I^+ \subset \widetilde{\mathfrak{n}}^+ be the ideal of 𝔫˜ +\widetilde{\mathfrak{n}}^+ generated by r +≔[e 1,e 1]r^+ \coloneqq [e_1, e_1]. It is spanned by the elements ad e i 1⋯ad e i k(r +)\mathrm{ad}_{e_{i_1}} \cdots \mathrm{ad}_{e_{i_k}}(r^+). Now [f 0,r +]=0[f_0, r^+] = 0, since [f 0,e 1]=0[f_0, e_1] = 0, and [f 1,r +]=2a 11e 1=0[f_1, r^+] = 2 a_{1 1} e_1 = 0, as computed in Example . By induction on kk, using

[f j,[e i,x]]=[[f j,e i],x]±[e i,[f j,x]] [f_j, [e_i, x]] \;=\; [[f_j, e_i], x] \pm [e_i, [f_j, x]]

with [f j,e i]∈𝔥[f_j, e_i] \in \mathfrak{h}, it follows that ad f j(I +)⊂I +\mathrm{ad}_{f_j}(I^+) \subset I^+. Since also ad 𝔥\mathrm{ad}_{\mathfrak{h}} and ad e i\mathrm{ad}_{e_i} preserve I +I^+, this is an ideal of all of 𝔤˜\widetilde{\mathfrak{g}}. Analogously, the ideal I −⊂𝔫˜ −I^- \subset \widetilde{\mathfrak{n}}^- generated by [f 1,f 1][f_1, f_1] is an ideal of 𝔤˜\widetilde{\mathfrak{g}}. Therefore

𝔤=𝔤˜/(I −⊕I +)=𝔫˜ −/I −⊕𝔥⊕𝔫˜ +/I +, \mathfrak{g} \;=\; \widetilde{\mathfrak{g}} \big/ \big(I^- \oplus I^+\big) \;=\; \widetilde{\mathfrak{n}}^- / I^- \;\oplus\; \mathfrak{h} \;\oplus\; \widetilde{\mathfrak{n}}^+ / I^+ \,,

and so 𝔫 +=𝔫˜ +/I +≃𝕃(e 0,e 1)/⟨[e 1,e 1]⟩\mathfrak{n}^+ = \widetilde{\mathfrak{n}}^+ / I^+ \simeq \mathbb{L}(e_0, e_1)/\langle [e_1, e_1] \rangle. Since AA is non-degenerate, 𝔤\mathfrak{g} has no center, and adjoining outer derivations does not affect 𝔫 +\mathfrak{n}^+, so that this holds for all the variants of Def. .

Item 2: Let J⊂𝔤J \subset \mathfrak{g} be an ideal with J∩𝔥=0J \cap \mathfrak{h} = 0. Since AA is non-degenerate, distinct elements of the root lattice define distinct weights, so that J=J −⊕J +J = J^- \oplus J^+ with J ±≔J∩𝔫 ±J^\pm \coloneqq J \cap \mathfrak{n}^\pm. Suppose that J −≠0J^- \neq 0, and pick a non-zero y∈J −y \in J^- of weight −β-\beta, with β\beta of minimal height. Then [e i,y]=0[e_i, y] = 0 for i∈{0,1}i \in \{0,1\}, by minimality and since J∩𝔥=0J \cap \mathfrak{h} = 0. Hence in the Verma module M(0)≃U(𝔫 −)M(0) \simeq U(\mathfrak{n}^-) of highest weight 00, the vector y⋅v 0y \cdot v_0 is a non-zero highest weight vector of weight −β-\beta.

Since AA is symmetric, 𝔤\mathfrak{g} carries an invariant even supersymmetric bilinear form with (h i|h j)=a ij(h_i \vert h_j) = a_{i j}, inducing (α i|α j)=a ij(\alpha_i \vert \alpha_j) = a_{i j} on the root lattice. The corresponding generalized Casimir operator (Kac 1990, §2.5, the construction carries over to the super case) commutes with 𝔤\mathfrak{g} and acts on highest weight vectors of weight λ\lambda by (λ|λ+2ρ)(\lambda \vert \lambda + 2\rho), where 2(ρ|α i)=(α i|α i)2(\rho \vert \alpha_i) = (\alpha_i \vert \alpha_i), hence here by (λ|λ)(\lambda \vert \lambda). As it acts by 00 on M(0)M(0), it follows that (β|β)=0(\beta \vert \beta) = 0. This is the argument of the Gabber-Kac theorem (Kac 1990, Thm. 9.11).

For β=nα 0+mα 1\beta = n \alpha_0 + m \alpha_1, the condition (β|β)=−2nm=0(\beta \vert \beta) = - 2 n m = 0 means that β=nα 0\beta = n \alpha_0 or β=mα 1\beta = m \alpha_1, with n,m≥2n, m \geq 2, since f i∈Jf_i \in J would imply h i=[e i,f i]∈J∩𝔥h_i = [e_i, f_i] \in J \cap \mathfrak{h}. But 𝔫 −nα 0 −=0\mathfrak{n}^-_{-n \alpha_0} = 0 and 𝔫 −mα 1 −=0\mathfrak{n}^-_{- m\alpha_1} = 0 for n,m≥2n, m \geq 2, because [f 0,f 0]=0[f_0, f_0] = 0 and [f 1,f 1]=0[f_1, f_1] = 0. This contradiction shows that J −=0J^- = 0, and analogously J +=0J^+ = 0.

Item 3: On the free Lie superalgebra 𝕃(a,b)\mathbb{L}(a, b) on an even generator aa and an odd generator bb, let DD be the odd derivation with D(a)=bD(a) = b and D(b)=0D(b) = 0. Then D 2=12[D,D]D^2 = \tfrac{1}{2}[D, D] is an even derivation vanishing on the generators, so that D 2=0D^2 = 0. Therefore the semidirect product S≔ℝe 1⋉𝕃(a,b)S \coloneqq \mathbb{R}e_1 \ltimes \mathbb{L}(a, b), with e 1e_1 odd, [e 1,e 1]=0[e_1, e_1] = 0 and [e 1,x]=D(x)[e_1, x] = D(x), is a Lie superalgebra. It is generated by e 1e_1 and aa, since b=[e 1,a]b = [e_1, a].

By item 1, there is a homomorphism ϕ:𝔫 +→S\phi \colon \mathfrak{n}^+ \to S with e 0↦ae_0 \mapsto a and e 1↦e 1e_1 \mapsto e_1. Conversely, let ψ:S→𝔫 +\psi \colon S \to \mathfrak{n}^+ send e 1↦e 1e_1 \mapsto e_1, and on 𝕃(a,b)\mathbb{L}(a,b) be given by a↦e 0a \mapsto e_0 and b↦[e 1,e 0]b \mapsto [e_1, e_0]. This is a homomorphism, because x↦ψ(Dx)x \mapsto \psi(D x) and x↦[e 1,ψ(x)]x \mapsto [e_1, \psi(x)] are odd derivations along ψ\psi which agree on the generators: on aa both give [e 1,e 0][e_1, e_0], and on bb both give zero, since [e 1,[e 1,e 0]]=12[[e 1,e 1],e 0]=0[e_1, [e_1, e_0]] = \tfrac{1}{2}[[e_1, e_1], e_0] = 0. Since ψ∘ϕ\psi \circ \phi and ϕ∘ψ\phi \circ \psi are the identity on generators, ϕ\phi is an isomorphism. This gives the first isomorphism, as [e 1,e 0]=−x 1[e_1, e_0] = - x_1.

The second isomorphism follows by Lazard elimination of e 0e_0 from 𝕃(e 0,x 1)\mathbb{L}(e_0, x_1), which gives 𝕃(e 0,x 1)≃ℝe 0⋉𝕃({(ad e 0) k(x 1)} k≥0)\mathbb{L}(e_0, x_1) \simeq \mathbb{R} e_0 \ltimes \mathbb{L}\big( \{ (\mathrm{ad}_{e_0})^k(x_1) \}_{k \geq 0} \big), using that 𝕃(e 0)=ℝe 0\mathbb{L}(e_0) = \mathbb{R} e_0 for even e 0e_0.

Item 4: The enveloping algebra of the quotient is the quotient of U(𝕃(e 0,e 1))=T(ℝe 0⊕ℝe 1)U\big(\mathbb{L}(e_0, e_1)\big) = T\big(\mathbb{R}e_0 \oplus \mathbb{R} e_1\big) by the two-sided ideal generated by [e 1,e 1]=2e 1e 1[e_1, e_1] = 2 e_1 e_1. The count of words is elementary, and the product formula is the PBW theorem for Lie superalgebras, U(𝔫 +)≃Sym(𝔫 even +)⊗∧(𝔫 odd +)U(\mathfrak{n}^+) \simeq \mathrm{Sym}(\mathfrak{n}^+_{even}) \otimes \wedge(\mathfrak{n}^+_{odd}) as graded vector spaces, using that the parity of nα 0+mα 1n \alpha_0 + m \alpha_1 is that of mm.

Remark

(IIA flux densities and roots)

Under the identification of the algebra of IIA flux densities with 𝔫 +\mathfrak{n}^+ (Greitz & Howe 2011b, §4.2), a flux density of degree 2n+m+12n + m + 1 corresponds to a root nα 0+mα 1n\alpha_0 + m\alpha_1, and it has even degree iff mm is odd. In the conventions above, the identification starts out as follows:

flux density
species
super Lie algebra elementBianchi identity
H 3H_3e 0e_0 (even)dH 3=0\mathrm{d} H_3 = 0
F 2F_2e 1e_1 (odd) [e 1,e 1]=0\;\;\;\; [e_1, e_1] = 0dF 2=0\mathrm{d} F_2 = 0
F 4F_4[e 0,e 1][e_0, e_1]dF 4=H 3F 2\mathrm{d} F_4 = H_3 F_2
F 6F_6[e 0,[e 0,e 1]]\big[e_0, [e_0, e_1]\big]dF 6=H 3F 4\mathrm{d} F_6 = H_3 F_4
H 7H_7[[e 0,e 1],[e 0,e 1]]=−[e 1,[e 0,[e 0,e 1]]]\big[[e_0, e_1], [e_0, e_1]\big] = -\Big[e_1, \big[e_0, [e_0, e_1]\big]\Big]dH 7=12F 4F 4−F 2F 6\mathrm{d} H_7 = \tfrac{1}{2} F_4 F_4 - F_2 F_6
F 8F_8[e 0,[e 0,[e 0,e 1]]]\Big[e_0,\big[e_0, [e_0, e_1]\big]\Big]dF 8=H 3F 6\mathrm{d} F_8 = H_3 F_6
H 9H_9[e 0,[[e 0,e 1],[e 0,e 1]]]=2[[e 0,e 1],[e 0,[e 0,e 1]]]=−23[e 1,[e 0,[e 0,[e 0,e 1]]]]\Big[e_0, \big[[e_0, e_1], [e_0, e_1]\big]\Big] = 2\Big[[e_0, e_1], \big[e_0, [e_0, e_1]\big]\Big] = -\tfrac{2}{3}\Big[e_1, \Big[e_0, \big[e_0, [e_0, e_1]\big]\Big]\Big]dH 9=H 3H 7+12F 4F 6−32F 2F 8\mathrm{d} H_9 = H_3 H_7 + \tfrac{1}{2} F_4 F_6 - \tfrac{3}{2} F_2 F_8
F 10F_{10}(ad e 0) 4(e 1)(\mathrm{ad}_{e_0})^4(e_1)dF 10=H 3F 8\mathrm{d} F_{10} = H_3 F_8
H 11H_{11}[e 0,[e 0,[[e 0,e 1],[e 0,e 1]]]]\Big[e_0, \Big[e_0, \big[[e_0, e_1], [e_0, e_1]\big]\Big]\Big]dH 11=H 3H 9−2F 2F 10+12F 4F 8\mathrm{d} H_{11} = H_3 H_9 - 2 F_2 F_{10} + \tfrac{1}{2} F_4 F_8
H′ 11H'_{11}[[e 0,[e 0,e 1]],[e 0,[e 0,e 1]]]\Big[\big[e_0, [e_0, e_1]\big], \big[e_0, [e_0, e_1]\big]\Big]dH′ 11=F 2F 10−F 4F 8+12F 6F 6\mathrm{d} H'_{11} = F_2 F_{10} - F_4 F_8 + \tfrac{1}{2} F_6 F_6
F 12F_{12}(ad e 0) 5(e 1)(\mathrm{ad}_{e_0})^5(e_1)dF 12=H 3F 10\mathrm{d} F_{12} = H_3 F_{10}
K 12K_{12}[[e 0,[e 0,e 1]],[[e 0,e 1],[e 0,e 1]]]=12[e 1,[[e 0,[e 0,e 1]],[e 0,[e 0,e 1]]]]\Big[\big[e_0, [e_0, e_1]\big], \big[[e_0, e_1], [e_0, e_1]\big]\Big] = \tfrac{1}{2}\Big[e_1, \Big[\big[e_0, [e_0, e_1]\big], \big[e_0, [e_0, e_1]\big]\Big]\Big]dK 12=−F 2H 11−2F 2H′ 11+F 4H 9−F 6H 7\mathrm{d} K_{12} = - F_2 H_{11} - 2 F_2 H'_{11} + F_4 H_9 - F_6 H_7

The two 11-fluxes reproduce the 2-parameter family of Greitz & Howe 2011b, (2.15), after conversion of conventions.

In terms of Prop. : the RR-fluxes F 2k+2F_{2k+2}, for k≥1k \geq 1, correspond to the free generators x kx_k, and all other flux densities except H 3H_3 and F 2F_2 correspond to iterated brackets of the x kx_k. In particular, the NS-fluxes beyond H 3H_3 correspond to the 2-fold brackets [x i,x j][x_i, x_j], whence the root nα 0+2α 1n \alpha_0 + 2 \alpha_1 has multiplicity ⌊n/2⌋\lfloor n/2 \rfloor. This reproduces the two series found by Greitz & Howe 2011b, §4.2: the (4n+3)(4n+3)-fluxes at roots 2nα 0+2α 12n\alpha_0 + 2\alpha_1 and the (4n+5)(4n+5)-fluxes at roots (2n+1)α 0+2α 1(2n+1)\alpha_0 + 2\alpha_1, both of multiplicity nn. The multiplicity 2 at n=2n = 2 matches the two independent 11-flux and 13-flux Bianchi identities found in the superspace analysis (Greitz & Howe 2011b, (2.15), (3.8)).

However, 𝔫 +\mathfrak{n}^+ also contains mm-fold brackets of the x kx_k for m≥3m \geq 3, which are not listed in Greitz & Howe 2011b. The first of these is the 12-flux K 12K_{12} at the root 4α 0+3α 14\alpha_0 + 3\alpha_1, in addition to the RR 12-flux F 12F_{12} at the root 5α 0+α 15 \alpha_0 + \alpha_1. By Prop. , item 4, the multiplicities mult(n,m)\mathrm{mult}(n, m) of the roots nα 0+mα 1n\alpha_0 + m\alpha_1 start out as follows:

n=0n = 0112233445566778899
m=0m = 000110000000000000000
m=1m = 111111111111111111111
m=2m = 200001111222233334444
m=3m = 300000000112233557799
m=4m = 40000000000112255881414
m=5m = 50000000000001133771414

Beyond the spacetime limit, most of these higher fluxes vanish in supergravity, but not all of them. The RR 12-flux F 12F_{12} is non-vanishing: its component with ten even and two odd superspace indices is a non-zero multiple of e −ϕΓ 10,2e^{-\phi}\, \Gamma_{10,2}, and its Bianchi identity dF 12=H 3F 10\mathrm{d} F_{12} = H_3 \, F_{10} is automatically solved (Greitz & Howe 2011b, (3.7)). The NS 13-fluxes vanish identically, but their Bianchi identities have non-vanishing terms on the right, whose sum vanishes by the cohomology argument. It has been speculated that such higher-degree forms may become non-zero once string-theoretic higher-order corrections are included (Greitz & Howe 2011b, §5).

Remark

(flux densities as closed 𝔫 +\mathfrak{n}^+-valued forms)

Grading 𝔫 +\mathfrak{n}^+ by the potential degree pp, so that in the IIA case p(α 0)=2p(\alpha_0) = 2 and p(α 1)=1p(\alpha_1) = 1, the super-parity of each root space is pp mod 2. This makes 𝔫 +\mathfrak{n}^+ a ℤ \mathbb{Z} -graded Lie algebra concentrated in positive degrees, hence an L-infinity algebra with only a binary bracket. Up to the sign convention 𝔽↦−𝔽\mathbb{F} \mapsto - \mathbb{F}, the combined Bianchi identity d𝔽=12[𝔽,𝔽]\mathrm{d} \mathbb{F} = \frac{1}{2}[\mathbb{F}, \mathbb{F}] then says that the flux densities constitute a closed 𝔫 + \mathfrak{n}^+ -valued differential form, hence a dg-algebra homomorphism

CE(𝔫 +)⟶Ω dR •(X) \mathrm{CE}\big(\mathfrak{n}^+\big) \longrightarrow \Omega^\bullet_{\mathrm{dR}}(X)

out of the Chevalley-Eilenberg algebra of 𝔫 +\mathfrak{n}^+, in which the generators dual to root spaces of potential degree pp sit in degree p+1p+1. This is the form in which Bianchi identities enter flux quantization.

For example, in the IIA case the root spaces with α 1\alpha_1-coefficient ≥2\geq 2 span an ideal in 𝔫 +\mathfrak{n}^+ (by Prop. these are the brackets of two or more of the x kx_k). The quotient by this ideal retains only H 3H_3 and the RR-fluxes F 2kF_{2k}, with their Bianchi identities dH 3=0\mathrm{d} H_3 = 0 and dF 2k+2=H 3F 2k\mathrm{d} F_{2k+2} = H_3 \, F_{2k}, which are those of the twisted Chern character in twisted K-theory.

Type IIB

Example

(form algebra of type IIB supergravity)

All flux densities of type IIB supergravity have odd degree, so that its Borcherds algebra is purely even. Its Cartan matrix is

(0 −1 −1 2), \left( \array{ 0 & -1 \\ -1 & 2 } \right) \,,

which gives the rank-2 Borcherds algebra previously considered by Slansky 1993 (Henry-Labordère, Julia & Paulot 2002, Greitz & Howe 2011b, §4.1). Its two simple roots play the following roles:

  • the real simple root α 1\alpha_1 generates a copy of 𝔰𝔩(2)\mathfrak{sl}(2), the (complexified) Lie algebra of the U-duality group SL(2,ℝ)SL(2,\mathbb{R}), with e 1e_1 associated with the axion;

  • the isotropic simple root α 0\alpha_0 corresponds to the lowest weight of the doublet of 3-fluxes.

Flux densities of degree 2n+12n+1 sit at level nn, the coefficient of α 0\alpha_0, and the algebra of flux densities is the part of 𝔫 +\mathfrak{n}^+ of positive level. Up to level 7 their 𝔰𝔩(2)\mathfrak{sl}(2)-content is

2,1,2,3,4+2,5+3+1,6+2⋅4+2⋅2. \mathbf{2} ,\;\; \mathbf{1} ,\;\; \mathbf{2} ,\;\; \mathbf{3} ,\;\; \mathbf{4} + \mathbf{2} ,\;\; \mathbf{5} + \mathbf{3} + \mathbf{1} ,\;\; \mathbf{6} + 2 \cdot \mathbf{4} + 2 \cdot \mathbf{2} \,.

This matches, including the multiplicities of the 15-fluxes, the consistent Bianchi identities found in superspace (Greitz & Howe 2011b, §3.1, §4.1).

Other dimensions

Analogous statements hold in other dimensions:

References

General

Original articles:

Textbook accounts:

  • Victor Kac: §11.13 in: Infinite dimensional Lie algebras, Cambridge University Press, 3rd edition (1990) [doi:10.1017/CBO9780511626234]

    (the first two editions do not contain the material on generalized Kac-Moody algebras)

  • Urmie Ray: Automorphic Forms and Lie Superalgebras, Algebra and Applications 5, Springer (2006) [doi:10.1007/978-1-4020-5010-7]

Introductions and reviews:

On the structure theory:

Monster Lie algebra and moonshine

See also:

Automorphic products

In string theory

As algebras of BPS states:

In the counting of dyons:

Symmetric products and second quantized strings:

In supergravity

On the Lie superalgebra structure encoded by the Bianchi identities of higher gauge field flux densities in supergravity (“superdualities”):

Identification of these as (truncations of) Borcherds superalgebras, and relation to U-duality and E11 (via mysterious duality):

The Borcherds algebra relevant for type IIB supergravity was first considered in:

The complete form spectra (including 10-form potentials) of type II supergravity, in components and in superspace:

On Borcherds algebras of Bianchi identities of supergravity flux densities in superspace, including form degrees beyond the spacetime dimension:

with input from superspace cohomology:

The analogous derivation of the form spectra from E 11 E_{11} :

Relation to tensor hierarchies in gauged supergravity:

Relation to exceptional generalized geometry:

including the induced L-infinity algebras:

See also:

Last revised on November 9, 2026 at 18:14:56. See the history of this page for a list of all contributions to it.