Introduction
Based on the deformation theory of complex structures going back to Kodaira and Spencer [1 ] , Kodaira–Spencer gravity is a topological field theory introduced by Bershadsky, Cecotti, Ooguri and Vafa (BCOV) as the closed string field theory of the so-called topological B-model [2 , 3 ] , whose open string counterpart is the holomorphic Chern–Simons theory of [4 ] . The perturbative quantization of BCOV, which has been formulated in the Batalin–Vilkovisky (BV) formalism by Costello and Li [5 , 6 ] , yields the higher genus amplitudes of topological string theory. (See also [7 ] for a proposal to obtain Kodaira–Spencer gravity from a twist of type IIB supergravity, which in turn features in ‘twisted holography’ [8 ] .)
The goal of this paper is to present a local action principle for BCOV theory or Kodaira–Spencer gravity. This theory is defined on a Calabi–Yau threefold X X X , and the gravity field is given by the complex structure deformation (also known as Beltrami differential), which in local coordinates is given by the mixed tensor A = A i ˉ j ( z , z ˉ ) d z ˉ i ˉ ⊗ ∂ ∂ z j A = A_{\bar{i}}^{j}(z,\bar{z}) \, \mathrm{d}\bar{z}^{\bar{i}} \otimes\frac{\partial}{\partial z^{j}} A = A i ˉ j ( z , z ˉ ) d z ˉ i ˉ ⊗ ∂ z j ∂ , where z z z and z ˉ \bar{z} z ˉ are holomorphic and antiholomorphic coordinates, respectively. The linearized field equations state that A A A is closed with respect to the antiholomorphic Dolbeault differential,
∂ ˉ A = 0. (1) \bar{\partial} A = 0 . \tag*{(1)} ∂ ˉ A = 0. (1)
These equations cannot be obtained from a local action principle for A A A , but one can write a non-local action upon imposing a constraint as follows. Any Calabi–Yau threefold carries a holomorphic ( 3 , 0 ) (3,0) ( 3 , 0 ) form with components ω i j k \omega_{ijk} ω ij k , which can be used to define the ( 2 , 1 ) (2,1) ( 2 , 1 ) form A i j k ˉ ′ = ω i j l A k ˉ l A'_{ij\bar{k}}=\omega_{ijl}A^{l}_{\bar{k}} A ij k ˉ ′ = ω ij l A k ˉ l . In terms of this we impose the constraint ∂ A ′ = 0 \partial A'=0 ∂ A ′ = 0 , which is equivalent to the original Beltrami differential being divergence-free in its holomorphic vector index, i.e., ∂ j A i ˉ j = 0 \partial_j A^{j}_{\bar{i}}=0 ∂ j A i ˉ j = 0 . One can then show that ∂ ˉ A ′ \bar{\partial}A' ∂ ˉ A ′ is ∂ \partial ∂ exact, i.e., ∂ ˉ A ′ = ∂ Ψ \bar{\partial}A'=\partial\Psi ∂ ˉ A ′ = ∂ Ψ for a ( 1 , 2 ) (1,2) ( 1 , 2 ) form Ψ \Psi Ψ . This defines the non-local operator ∂ − 1 ∂ ˉ A ′ : = Ψ \partial^{-1}\bar{\partial}A':=\Psi ∂ − 1 ∂ ˉ A ′ := Ψ , in terms of which one can write the action
S k i n = 1 2 ∫ X A ′ ∧ ∂ − 1 ∂ ˉ A ′ . (2) S_{\mathrm{kin}}=\frac{1}{2}\int_X A'\wedge\partial^{-1}\bar{\partial}A'. \tag*{(2)} S kin = 2 1 ∫ X A ′ ∧ ∂ − 1 ∂ ˉ A ′ . (2)
Variation with respect to A ′ A' A ′ yields ∂ ˉ A ′ = 0 \bar{\partial}A'=0 ∂ ˉ A ′ = 0 , which, owing to ∂ ˉ \bar{\partial} ∂ ˉ commuting with the holomorphic ω \omega ω , in turn implies the original Kodaira–Spencer gravity equations (1) .
In this paper we use that the above action can be made local by solving the constraint ∂ A ′ = 0 \partial A'=0 ∂ A ′ = 0 in terms of a new ( 1 , 1 ) (1,1) ( 1 , 1 ) form field e = e i j ˉ ( z , z ˉ ) d z i ∧ d z j ˉ e=e_{i\bar{j}}(z,\bar{z})dz^i\wedge dz^{\bar{j}} e = e i j ˉ ( z , z ˉ ) d z i ∧ d z j ˉ :
A ′ = ∂ e ⟺ A j ˉ k = ω k l i ∂ l e i j ˉ , (3) A'=\partial e \qquad\Longleftrightarrow\qquad A^{k}_{\bar{j}}=\omega^{kli}\partial_l e_{i\bar{j}}, \tag*{(3)} A ′ = ∂ e ⟺ A j ˉ k = ω k l i ∂ l e i j ˉ , (3)
as already observed by BCOV (see footnote 23 in [3 ] ). In order to see that this makes the quadratic term local we note with (1.3) that ∂ ˉ A ′ = − ∂ ∂ ˉ e \bar{\partial}A'=-\partial\bar{\partial}e ∂ ˉ A ′ = − ∂ ∂ ˉ e , so we can write ∂ − 1 ∂ ˉ A ′ = − ∂ ˉ e \partial^{-1}\bar{\partial}A'=-\bar{\partial}e ∂ − 1 ∂ ˉ A ′ = − ∂ ˉ e , and the kinetic term (1.2) localizes to the two-derivative action
S k i n = − 1 2 ∫ X ∂ e ∧ ∂ ˉ e . (4) S_{\mathrm{kin}}=-\frac{1}{2}\int_X\partial e\wedge\bar{\partial}e. \tag*{(4)} S kin = − 2 1 ∫ X ∂ e ∧ ∂ ˉ e . (4)
Furthermore, Kodaira–Spencer gravity has a cubic interaction term in terms of A A A or, equivalently, A ′ A' A ′ that is local and in fact carries no derivatives. We can hence write a local interaction term in terms of e e e by simply replacing A ′ A' A ′ according to (1.3), which yields cubic interaction terms with three derivatives.
It is thus guaranteed that the entire action is local, but the interesting question is how the gauge symmetries including diffeomorphisms are realized in this formulation. Indeed, writing (1.3) introduces a gauge redundancy since from this relation e e e is only well-defined up to shifts e → e + ∂ λ ˉ e\to e+\partial\bar{\lambda} e → e + ∂ λ ˉ , with λ ˉ = λ ˉ i ˉ ( z , z ˉ ) d z i ˉ \bar{\lambda}=\bar{\lambda}_{\bar{i}}(z,\bar{z})dz^{\bar{i}} λ ˉ = λ ˉ i ˉ ( z , z ˉ ) d z i ˉ a ( 0 , 1 ) (0,1) ( 0 , 1 ) form. Furthermore, standard Kodaira–Spencer gravity is invariant under volume-preserving holomorphic diffeomorphisms, with a parameter that can be encoded in a ( 1 , 0 ) (1,0) ( 1 , 0 ) form λ = λ i ( z , z ˉ ) d z i \lambda=\lambda_i(z,\bar{z})dz^i λ = λ i ( z , z ˉ ) d z i , both acting on e e e as1
δ e = ∂ λ ˉ + ∂ ˉ λ ⟺ δ e i j ˉ = ∂ i λ ˉ j ˉ − ∂ ˉ j ˉ λ i , (5) \delta e=\partial\bar{\lambda}+\bar{\partial}\lambda\qquad\Longleftrightarrow\qquad\delta e_{i\bar{j}}=\partial_i\bar{\lambda}_{\bar{j}}-\bar{\partial}_{\bar{j}}\lambda_i, \tag*{(5)} δ e = ∂ λ ˉ + ∂ ˉ λ ⟺ δ e i j ˉ = ∂ i λ ˉ j ˉ − ∂ ˉ j ˉ λ i , (5)
where λ \lambda λ and λ ˉ \bar{\lambda} λ ˉ are independent parameters (and not complex conjugate to each other). The standard diffeomorphism action is with respect to the vector parameter
ϑ ( λ ) i = ω i j k ∂ j λ k , (6) \vartheta(\lambda)^i=\omega^{ijk}\partial_j\lambda_k, \tag*{(6)} ϑ ( λ ) i = ω ij k ∂ j λ k , (6)
which is divergence-free and hence generates volume-preserving diffeomorphisms. The non-linear gauge algebra turns out to be an intriguing higher L ∞ L_\infty L ∞ and Leibniz algebra structure. based on a Courant algebroid with a non-trivial three-bracket. Remarkably, precisely the same Leibniz algebras have appeared in gauged supergravity [10 , 11 , 12 ] and double and exceptional field theory [13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 ] , with the ‘anchor’ or ‘embedding tensor’ being given by analogues of (6) [21 , 22 , 23 , 24 , 25 , 26 ] , see [27 ] for a closely related precursor. Indeed, in our formulation the perturbative gravity field e i j ˉ e_{i\bar{j}} e i j ˉ is identical to the generalized metric fluctuation of double field theory, with the same linearized gauge transformations [13 , 14 , 15 , 16 ] . Moreover, reducing to the three-dimensional (3D) real slice z = z ˉ z=\bar{z} z = z ˉ and decomposing e e e into its symmetric and antisymmetric parts, the action (4) yields the Fierz–Pauli action for massless spin-2 plus the Kalb–Ramond action for a B-field. At higher order we obtain parity-odd higher-derivative couplings. As in double field theory the gauge algebra is a Courant algebroid, except that here no level-matching or ‘section’ constraint, which would halve the number of dimensions, needs to be imposed. In this sense, Kodaira–Spencer gravity is an unconstrained double field theory.
The rest of this paper is organized as follows. In section 2 we give a self-contained review of Kodaira–Spencer gravity, introducing the Dolbeault bicomplex, the Kodaira–Spencer equations and the non-local action. The local reformulation of the action principle sketched above is developed in section 3. We begin by introducing the Leibniz algebra underlying the Roytenberg–Weinstein construction of the L ∞ L_\infty L ∞ algebra of Courant algebroids [28 ] , which will form part of the (higher) gauge algebra. Then we define a cyclic L ∞ L_\infty L ∞ algebra that encodes, in the sense of [29 , 30 ] , the full local reformulation of Kodaira–Spencer gravity and which is defined on a certain self-dual or ‘folded’ subspace of the Dolbeault complex. This cyclic L ∞ L_\infty L ∞ algebra defines the complete BV master action, which due to the higher gauge algebra includes quartic couplings in the ghost sector. In subsection 3.4 we reduce the theory to a real 3D slice and relate it to the Fierz–Pauli theory and to an exotic deformation thereof found by Boulanger and Gualtieri [31 ] . In section 4 we give a ‘double copy’ interpretation of the L ∞ L_\infty L ∞ algebra that is analogous, though different in important respects, to the homotopy algebraic double copy for which there is a growing literature [32 , 33 , 34 , 35 , 36 ] (see in particular [37 ] for a related double copy discussion of Kodaira–Spencer gravity). Conclusions and an outlook are in section 5, where we also compare with the recent works [38 ] and [39 ] , which relate Kodaira–Spencer gravity and its real analogue to three-dimensional gravity. In the bulk of this paper we assume that every ∂ ˉ \bar{\partial} ∂ ˉ -closed ( 2 , 1 ) (2,1) ( 2 , 1 ) -form on X X X is ∂ ˉ \bar{\partial} ∂ ˉ -exact, so that ∂ ˉ A ′ = 0 \bar{\partial}A'=0 ∂ ˉ A ′ = 0 can indeed be solved by (3) . This still captures several interesting cases, including X = C 3 X=\mathbb{C}^{3} X = C 3 , but in subsection 4.3 we show that it is relatively straightforward to relax this assumption by including the moduli corresponding to the harmonic pieces in the cohomology of a compact Calabi–Yau threefold.
Non-local action for Kodaira–Spencer gravity
In this section we review standard Kodaira–Spencer gravity. To keep the paper self-contained, in the first subsection we recall the basics of complex three-manifolds and introduce the Dolbeault complex of differential forms. The equations of motion of Kodaira–Spencer gravity are defined in the second subsection, based on the complex of antiholomorphic forms taking values in the holomorphic polyvectors. The final subsection introduces the non-local action.
Generalities
Let X X X be a Calabi–Yau threefold. The reader may think of X = C 3 X = \mathbb{C}^{3} X = C 3 , but the following results will be valid for any Calabi–Yau threefold. We can view C 3 \mathbb{C}^{3} C 3 as the real six-dimensional manifold R 6 \mathbb{R}^{6} R 6 with coordinates
X M = ( x i , y i ) , i = 1 , 2 , 3. (7) X^{M} = (x^{i}, y^{i}), \quad i = 1,2,3 . \tag*{(7)} X M = ( x i , y i ) , i = 1 , 2 , 3. (7)
We can alternatively introduce complex coordinates
z i = x i + i y i , z ˉ i ˉ = x i − i y i , (8) z^{i} = x^{i} + i y^{i}, \qquad\bar{z}^{\bar{i}} = x^{i} - i y^{i}, \tag*{(8)} z i = x i + i y i , z ˉ i ˉ = x i − i y i , (8)
where again i = 1 , 2 , 3 i = 1,2,3 i = 1 , 2 , 3 , and an i i i that does not appear as an index is a complex unit with i 2 = − 1 i^{2} = -1 i 2 = − 1 . These formulas motivate the definition of the following differential operators
∂ i : = ∂ ∂ z i = 1 2 ( ∂ ∂ x i − i ∂ ∂ y i ) , ∂ ˉ i ˉ : = ∂ ∂ z ˉ i ˉ = 1 2 ( ∂ ∂ x i + i ∂ ∂ y i ) , (9) \partial_{i} := \frac{\partial}{\partial z^{i}} = \frac{1}{2}\left(\frac{\partial}{\partial x^{i}} - i\frac{\partial}{\partial y^{i}}\right), \qquad\bar{\partial}_{\bar{i}} := \frac{\partial}{\partial\bar{z}^{\bar{i}}} = \frac{1}{2}\left(\frac{\partial}{\partial x^{i}} + i\frac{\partial}{\partial y^{i}}\right), \tag*{(9)} ∂ i := ∂ z i ∂ = 2 1 ( ∂ x i ∂ − i ∂ y i ∂ ) , ∂ ˉ i ˉ := ∂ z ˉ i ˉ ∂ = 2 1 ( ∂ x i ∂ + i ∂ y i ∂ ) , (9)
which obey ∂ i z j = δ i j = ∂ ˉ i ˉ z ˉ j ˉ \partial_{i}z^{j} = \delta^{j}_{i} = \bar{\partial}_{\bar{i}}\bar{z}^{\bar{j}} ∂ i z j = δ i j = ∂ ˉ i ˉ z ˉ j ˉ and ∂ i z ˉ j ˉ = 0 = ∂ ˉ i ˉ z j \partial_{i}\bar{z}^{\bar{j}} = 0 = \bar{\partial}_{\bar{i}}z^{j} ∂ i z ˉ j ˉ = 0 = ∂ ˉ i ˉ z j .
Next we define the space Ω p , q \Omega^{p,q} Ω p , q of ( p , q ) (p,q) ( p , q ) -forms (the Dolbeault complex). A ( p , q ) (p,q) ( p , q ) -form can be written as
α = 1 p ! q ! α i 1 ⋯ i p ȷ ˉ 1 ⋯ ȷ ˉ q ( z , z ˉ ) d z i 1 ∧ ⋯ ∧ d z i p ∧ d z ˉ ȷ ˉ 1 ∧ ⋯ ∧ d z ˉ ȷ ˉ q , (10) \alpha= \frac{1}{p!q!}\alpha_{i_{1}\cdots i_{p}\bar{\jmath}_{1}\cdots\bar{\jmath}_{q}}(z,\bar{z})\, dz^{i_{1}}\wedge\cdots\wedge dz^{i_{p}}\wedge d\bar{z}^{\bar{\jmath}_{1}}\wedge\cdots\wedge d\bar{z}^{\bar{\jmath}_{q}}, \tag*{(10)} α = p ! q ! 1 α i 1 ⋯ i p ˉ 1 ⋯ ˉ q ( z , z ˉ ) d z i 1 ∧ ⋯ ∧ d z i p ∧ d z ˉ ˉ 1 ∧ ⋯ ∧ d z ˉ ˉ q , (10)
whose components are separately antisymmetric in the first group of p p p indices and the last group of q q q indices. The familiar calculus of differential forms of a real manifold generalizes as follows: The de Rham differential is replaced by two differentials ∂ \partial ∂ and ∂ ˉ \bar{\partial} ∂ ˉ that act on functions as ∂ f = ∂ i f d z i \partial f = \partial_{i}f\, dz^{i} ∂ f = ∂ i f d z i and ∂ ˉ f = ∂ ˉ i ˉ f d z ˉ i ˉ \bar{\partial}f = \bar{\partial}_{\bar{i}}f\, d\bar{z}^{\bar{i}} ∂ ˉ f = ∂ ˉ i ˉ f d z ˉ i ˉ and on arbitrary ( p , q ) (p,q) ( p , q ) forms as
∂ = d z i ∧ ∂ i : Ω p , q → Ω p + 1 , q , ∂ ˉ = d z ˉ i ˉ ∧ ∂ ˉ i ˉ : Ω p , q → Ω p , q + 1 . (11) \partial= dz^{i}\wedge\partial_{i} : \Omega^{p,q}\to\Omega^{p+1,q}, \qquad\bar{\partial} = d\bar{z}^{\bar{i}}\wedge\bar{\partial}_{\bar{i}} : \Omega^{p,q}\to\Omega^{p,q+1}. \tag*{(11)} ∂ = d z i ∧ ∂ i : Ω p , q → Ω p + 1 , q , ∂ ˉ = d z ˉ i ˉ ∧ ∂ ˉ i ˉ : Ω p , q → Ω p , q + 1 . (11)
These operators square to zero just as the familiar de Rham differential does. Moreover, it follows by a straightforward computation that ∂ \partial ∂ and ∂ ˉ \bar{\partial} ∂ ˉ anticommute, so that we have in total
∂ 2 = 0 , ∂ ˉ 2 = 0 , ∂ ∂ ˉ + ∂ ˉ ∂ = 0. (12) \partial^{2} = 0, \qquad\bar{\partial}^{2} = 0, \qquad\partial\bar{\partial} + \bar{\partial}\partial= 0 . \tag*{(12)} ∂ 2 = 0 , ∂ ˉ 2 = 0 , ∂ ∂ ˉ + ∂ ˉ ∂ = 0. (12)
These relations in turn imply that there is a total differential
d = ∂ + ∂ ˉ , (13) d = \partial+ \bar{\partial}, \tag*{(13)} d = ∂ + ∂ ˉ , (13)
that squares to zero, d 2 = 0 d^{2} = 0 d 2 = 0 . The total space of ( p , q ) (p,q) ( p , q ) forms denoted by Ω ∙ , ∙ \Omega^{\bullet,\bullet} Ω ∙ , ∙ then forms a Dolbeault bicomplex : a bigraded space with two anticommuting derivations. Moreover, ( Ω ∙ , ∙ , ∧ , ∂ , ∂ ˉ ) (\Omega^{\bullet,\bullet}, \wedge, \partial, \bar{\partial}) ( Ω ∙ , ∙ , ∧ , ∂ , ∂ ˉ ) carries a bigraded-commutative algebra structure given by the wedge product ∧ \wedge ∧ .
We will frequently use the so-called ∂ ∂ ˉ \partial\bar{\partial} ∂ ∂ ˉ -lemma, which relates the cohomologies of ∂ \partial ∂ , ∂ ˉ \bar{\partial} ∂ ˉ and d d d . On a compact Kähler manifold, and hence on a compact Calabi–Yau threefold, it states that for a form α \alpha α that is both ∂ \partial ∂ -closed and ∂ ˉ \bar{\partial} ∂ ˉ -closed the following conditions are equivalent: α \alpha α is d d d -exact, α \alpha α is ∂ \partial ∂ -exact, α \alpha α is ∂ ˉ \bar{\partial} ∂ ˉ -exact, and α = ∂ ∂ ˉ β \alpha= \partial\bar{\partial}\beta α = ∂ ∂ ˉ β for some β \beta β [40 , 41 ] . The local version holds without compactness: locally, every d d d -closed ( p , q ) (p,q) ( p , q ) -form with p , q ≥ 1 p,q \ge1 p , q ≥ 1 is of the form ∂ ∂ ˉ β \partial\bar{\partial}\beta ∂ ∂ ˉ β .
In the following the holomorphic polyvector fields will also play an important role. These are fields with antisymmetric upper holomorphic indices whose space we denote by ∧ p T 1 , 0 \wedge^{p}T^{1,0} ∧ p T 1 , 0 . In order to relate forms to polyvectors we will assume that there is a nowhere vanishing holomorphic ( 3 , 0 ) (3,0) ( 3 , 0 ) -form ω \omega ω . In local coordinates we write
ω = 1 3 ! ω i j k d z i ∧ d z j ∧ d z k ∈ Ω ( 3 , 0 ) (14) \omega= \frac{1}{3!}\omega_{ijk} dz^i \wedge dz^j \wedge dz^k \in\Omega^{(3,0)} \tag*{(14)} ω = 3 ! 1 ω ij k d z i ∧ d z j ∧ d z k ∈ Ω ( 3 , 0 ) (14)
where the ω i j k \omega_{ijk} ω ij k are totally antisymmetric. By degree reasons this form obeys ∂ ω = 0 \partial\omega= 0 ∂ ω = 0 , and we will also assume ∂ ˉ ω = 0 \bar{\partial}\omega= 0 ∂ ˉ ω = 0 . We further introduce the dual trivector
ω − 1 = 1 3 ! ω i j k ∂ i ∧ ∂ j ∧ ∂ k ∈ ∧ 3 T 1 , 0 (15) \omega^{-1} = \frac{1}{3!}\omega^{ijk} \partial_i \wedge\partial_j \wedge\partial_k \in\wedge^3 T^{1,0} \tag*{(15)} ω − 1 = 3 ! 1 ω ij k ∂ i ∧ ∂ j ∧ ∂ k ∈ ∧ 3 T 1 , 0 (15)
whose components ω i j k \omega^{ijk} ω ij k are also totally antisymmetric. For an antiholomorphic form taking values in the holomorphic polyvectors we can write
β = 1 p ! q ! β ı ˉ 1 … ı ˉ q j 1 … j p ( z , z ˉ ) d z ˉ ı ˉ 1 ∧ ⋯ ∧ d z ˉ ı ˉ q ⊗ ∂ j 1 ∧ ⋯ ∧ ∂ j p ∈ Ω 0 , q ( ∧ p T 1 , 0 ) (16) \beta= \frac{1}{p!q!}\beta_{\bar{\imath}_1\ldots\bar{\imath}_q}^{j_1\ldots j_p}(z,\bar{z})d\bar{z}^{\bar{\imath}_1}\wedge\cdots\wedge d\bar{z}^{\bar{\imath}_q}\otimes\partial_{j_1}\wedge\cdots\wedge\partial_{j_p}\in\Omega^{0,q}(\wedge^pT^{1,0}) \tag*{(16)} β = p ! q ! 1 β ˉ 1 … ˉ q j 1 … j p ( z , z ˉ ) d z ˉ ˉ 1 ∧ ⋯ ∧ d z ˉ ˉ q ⊗ ∂ j 1 ∧ ⋯ ∧ ∂ j p ∈ Ω 0 , q ( ∧ p T 1 , 0 ) (16)
i.e., in local coordinates such a tensor is a mixed tensor, with unbarred upper indices and barred lower indices. Given the nowhere vanishing holomorphic ( 3 , 0 ) (3,0) ( 3 , 0 ) -form ω \omega ω we have an isomorphism mapping β ∈ Ω 0 , q ( ∧ p T 1 , 0 ) \beta\in\Omega^{0,q}(\wedge^pT^{1,0}) β ∈ Ω 0 , q ( ∧ p T 1 , 0 ) to β ⌟ ω ∈ Ω 3 − p , q \beta\lrcorner\omega\in\Omega^{3-p,q} β ┘ ω ∈ Ω 3 − p , q , which is given by the contraction of the upper (polyvector) indices with ω \omega ω ,
β ⌟ ω = 1 p ! ( 3 − p ) ! q ! β ı ˉ 1 … ı ˉ q j 1 … j p ω j 1 … j p k 1 … k 3 − p d z k 1 ∧ ⋯ ∧ d z k 3 − p ∧ d z ˉ ı ˉ 1 ∧ ⋯ ∧ d z ˉ ı ˉ q (17) \beta\lrcorner\omega= \frac{1}{p!(3-p)!q!}\beta_{\bar{\imath}_1\ldots\bar{\imath}_q}^{j_1\ldots j_p}\omega_{j_1\ldots j_p k_1\ldots k_{3-p}}dz^{k_1}\wedge\cdots\wedge dz^{k_{3-p}}\wedge d\bar{z}^{\bar{\imath}_1}\wedge\cdots\wedge d\bar{z}^{\bar{\imath}_q} \tag*{(17)} β ┘ ω = p ! ( 3 − p )! q ! 1 β ˉ 1 … ˉ q j 1 … j p ω j 1 … j p k 1 … k 3 − p d z k 1 ∧ ⋯ ∧ d z k 3 − p ∧ d z ˉ ˉ 1 ∧ ⋯ ∧ d z ˉ ˉ q (17)
Kodaira–Spencer theory
We are now ready to define Kodaira–Spencer theory. It is defined on the space of antiholomorphic forms taking values in the holomorphic polyvector fields,
P V = ⨁ i , j P V i , j , P V i , j = Ω 0 , j ( ∧ i T 1 , 0 ) (18) \mathrm{PV} = \bigoplus_{i,j}\mathrm{PV}^{i,j},\qquad\mathrm{PV}^{i,j}=\Omega^{0,j}(\wedge^iT^{1,0}) \tag*{(18)} PV = i , j ⨁ PV i , j , PV i , j = Ω 0 , j ( ∧ i T 1 , 0 ) (18)
where the degree is given by i + j i+j i + j [5 ] . This space carries the wedge product ∧ \wedge ∧ , under which it is graded commutative with respect to the total degree ∣ β ∣ = i + j |\beta|=i+j ∣ β ∣ = i + j of β ∈ P V i , j \beta\in\mathrm{PV}^{i,j} β ∈ PV i , j . Here both the antiholomorphic one-forms d z ˉ ı ˉ d\bar{z}^{\bar{\imath}} d z ˉ ˉ and the vector fields ∂ j \partial_j ∂ j are treated as odd, so that the wedge product of β ∈ P V p , q \beta\in\mathrm{PV}^{p,q} β ∈ PV p , q and γ ∈ P V p ′ , q ′ \gamma\in\mathrm{PV}^{p',q'} γ ∈ PV p ′ , q ′ is computed by moving the p p p vector fields of β \beta β past the q ′ q' q ′ one-forms of γ \gamma γ , which gives a sign ( − 1 ) p q ′ (-1)^{pq'} ( − 1 ) p q ′ . It further carries the Dolbeault differential ∂ ˉ = d z ˉ ı ˉ ∧ ∂ ˉ ı ˉ \bar{\partial}=d\bar{z}^{\bar{\imath}}\wedge\bar{\partial}_{\bar{\imath}} ∂ ˉ = d z ˉ ˉ ∧ ∂ ˉ ˉ , which acts from the left, and the divergence operator div ω : P V p , q → P V p − 1 , q \operatorname{div}_{\omega}:\mathrm{PV}^{p,q}\to\mathrm{PV}^{p-1,q} div ω : PV p , q → PV p − 1 , q , which contracts the first upper index with a holomorphic derivative.2
( div ω β ) ı ˉ 1 … ı ˉ q j 2 … j p = ( − 1 ) q ∂ j β ı ˉ 1 … ı ˉ q j j 2 … j p (19) (\operatorname{div}_{\omega}\beta)_{\bar{\imath}_1\ldots\bar{\imath}_q}^{j_2\ldots j_p}=(-1)^q\partial_j\beta_{\bar{\imath}_1\ldots\bar{\imath}_q}^{jj_2\ldots j_p} \tag*{(19)} ( div ω β ) ˉ 1 … ˉ q j 2 … j p = ( − 1 ) q ∂ j β ˉ 1 … ˉ q j j 2 … j p (19)
in terms of the components defined above. The sign ( − 1 ) q (-1)^q ( − 1 ) q arises because div ω \operatorname{div}_{\omega} div ω acts from the left and hence has to move past the q q q antiholomorphic one-forms. Under the isomorphism with forms the divergence corresponds to ∂ \partial ∂ in the sense that
( div ω β ) ⌟ ω = ( − 1 ) ∣ β ∣ + 1 ∂ ( β ⌟ ω ) . (20) (\operatorname{div}_{\omega}\beta)\lrcorner\omega=(-1)^{|\beta|+1}\partial(\beta\lrcorner\omega). \tag*{(20)} ( div ω β ) ┘ ω = ( − 1 ) ∣ β ∣ + 1 ∂ ( β ┘ ω ) . (20)
div ω \operatorname{div}_{\omega} div ω squares to zero and is a second-order operator for ∧ \wedge ∧ . This means that the triple ( P V , ∧ , div ω ) (\mathrm{PV},\wedge,\operatorname{div}_{\omega}) ( PV , ∧ , div ω ) is a BV algebra. This implies that the Schouten–Nijenhuis bracket
{ β , γ } : = ( − 1 ) β ( div ω ( β ∧ γ ) − div ω β ∧ γ − ( − 1 ) β β ∧ div ω γ ) (21) \{\beta,\gamma\}:=(-1)^{\beta}\left(\operatorname{div}_{\omega}(\beta\wedge\gamma)-\operatorname{div}_{\omega}\beta\wedge\gamma-(-1)^{\beta}\beta\wedge\operatorname{div}_{\omega}\gamma\right) \tag*{(21)} { β , γ } := ( − 1 ) β ( div ω ( β ∧ γ ) − div ω β ∧ γ − ( − 1 ) β β ∧ div ω γ ) (21)
where in exponents we write β \beta β for ∣ β ∣ |\beta| ∣ β ∣ , is graded (anti-)symmetric,
{ β , γ } = − ( − 1 ) ( β + 1 ) ( γ + 1 ) { γ , β } (22) \{\beta,\gamma\}=-(-1)^{(\beta+1)(\gamma+1)}\{\gamma,\beta\} \tag*{(22)} { β , γ } = − ( − 1 ) ( β + 1 ) ( γ + 1 ) { γ , β } (22)
and satisfies the graded Jacobi identity exactly,
{ β , { γ , α } } = { { β , γ } , α } + ( − 1 ) ( β + 1 ) ( γ + 1 ) { γ , { β , α } } . (23) \{\beta,\{\gamma,\alpha\}\}=\{\{\beta,\gamma\},\alpha\}+(-1)^{(\beta+1)(\gamma+1)}\{\gamma,\{\beta,\alpha\}\}. \tag*{(23)} { β , { γ , α }} = {{ β , γ } , α } + ( − 1 ) ( β + 1 ) ( γ + 1 ) { γ , { β , α }} . (23)
Moreover, ∂ ˉ \bar{\partial} ∂ ˉ is a derivation of the bracket,
∂ ˉ { α , β } = { ∂ ˉ α , β } + ( − 1 ) ∣ α ∣ + 1 { α , ∂ ˉ β } . (24) \bar{\partial}\{\alpha,\beta\}=\{\bar{\partial}\alpha,\beta\}+(-1)^{|\alpha|+1}\{\alpha,\bar{\partial}\beta\}. \tag*{(24)} ∂ ˉ { α , β } = { ∂ ˉ α , β } + ( − 1 ) ∣ α ∣ + 1 { α , ∂ ˉ β } . (24)
These relations make in particular ( P V , ∧ , { ⋅ , ⋅ } ) (\mathrm{PV},\wedge,\{\cdot,\cdot\}) ( PV , ∧ , { ⋅ , ⋅ }) a Gerstenhaber algebra [42 ] .
In order to verify (2.18) explicitly one has to use that ∂ ˉ \bar{\partial} ∂ ˉ is a derivation of degree one of the wedge product, ∂ ˉ ( β ∧ γ ) = ∂ ˉ β ∧ γ + ( − 1 ) ∣ β ∣ β ∧ ∂ ˉ γ \bar{\partial}(\beta\wedge\gamma)=\bar{\partial}\beta\wedge\gamma+(-1)^{|\beta|}\beta\wedge\bar{\partial}\gamma ∂ ˉ ( β ∧ γ ) = ∂ ˉ β ∧ γ + ( − 1 ) ∣ β ∣ β ∧ ∂ ˉ γ and that ∂ ˉ \bar{\partial} ∂ ˉ and div ω \operatorname{div}_{\omega} div ω anticommute,
∂ ˉ div ω + div ω ∂ ˉ = 0. (25) \bar{\partial}\operatorname{div}_{\omega}+\operatorname{div}_{\omega}\bar{\partial}=0. \tag*{(25)} ∂ ˉ div ω + div ω ∂ ˉ = 0. (25)
This relation in turn can be quickly verified in components or in index-free form using (2.14), for which one has to use that for β ∈ P V p , q \beta\in\mathrm{PV}^{p,q} β ∈ PV p , q
( ∂ ˉ β ) ⌟ ω = ( − 1 ) p + 1 ∂ ˉ ( β ⌟ ω ) , (26) (\bar{\partial}\beta)\lrcorner\omega=(-1)^{p+1}\bar{\partial}(\beta\lrcorner\omega), \tag*{(26)} ( ∂ ˉ β ) ┘ ω = ( − 1 ) p + 1 ∂ ˉ ( β ┘ ω ) , (26)
since ω \omega ω is holomorphic, so that ∂ ˉ \bar{\partial} ∂ ˉ acts only on the components of β \beta β .
We are now ready to define Kodaira–Spencer gravity. To this end we note that the subspace ( P V , ∂ ˉ ) (\mathrm{PV},\bar{\partial}) ( PV , ∂ ˉ ) forms a differential graded (dg) Lie algebra, which implies that we can postulate Maurer–Cartan equations as the field equations. These define the Kodaira–Spencer equations for the gravity field A = A l ˉ j ( z , z ˉ ) d z l ˉ ⊗ ∂ ∂ z j ∈ P V 1 , 1 A=A_{\bar l}^{j}(z,\bar z)\,dz^{\bar l}\otimes\frac{\partial}{\partial z^{j}}\in\mathrm{PV}^{1,1} A = A l ˉ j ( z , z ˉ ) d z l ˉ ⊗ ∂ z j ∂ ∈ PV 1 , 1 :
∂ ˉ A + 1 2 { A , A } = 0. (27) \bar{\partial}A+\frac{1}{2}\{A,A\}=0. \tag*{(27)} ∂ ˉ A + 2 1 { A , A } = 0. (27)
These equations are gauge covariant under
δ ξ A = ∂ ˉ ξ + { A , ξ } , ξ ∈ P V 1 , 0 , (28) \delta_{\xi}A=\bar{\partial}\xi+\{A,\xi\},\qquad\xi\in\mathrm{PV}^{1,0}, \tag*{(28)} δ ξ A = ∂ ˉ ξ + { A , ξ } , ξ ∈ PV 1 , 0 , (28)
with a vector gauge parameter ξ \xi ξ . Covariance follows from ∂ ˉ 2 = 0 \bar{\partial}^{2}=0 ∂ ˉ 2 = 0 , the Leibniz rule (2.18) and the Jacobi identity (2.17).
Non-local action
There is no local action for A A A that implies the Kodaira–Spencer equations (2.21), but one can define a non-local action under an assumption as follows: We set
A ′ = A ⌟ ω ∈ Ω 2 , 1 ⟺ A i j l ˉ ′ = ω i j k A l ˉ k , (29) A'=A\lrcorner\omega\in\Omega^{2,1}\qquad\Longleftrightarrow\qquad A'_{ij\bar l}=\omega_{ijk}A_{\bar l}^{k}, \tag*{(29)} A ′ = A ┘ ω ∈ Ω 2 , 1 ⟺ A ij l ˉ ′ = ω ij k A l ˉ k , (29)
and then assume that this ( 2 , 1 ) (2,1) ( 2 , 1 ) form is ∂ \partial ∂ closed,
∂ A ′ = 0. (30) \partial A' = 0 . \tag*{(30)} ∂ A ′ = 0. (30)
This is equivalent to the divergence div ω A \operatorname{div}_{\omega} A div ω A being zero (in local coordinates ∂ j A i ˉ j = 0 \partial_j A_{\bar{i}}^j = 0 ∂ j A i ˉ j = 0 ). Indeed, by (20) we have ∂ A ′ = ∂ ( A ⌟ ω ) = − ( div ω A ) ⌟ ω = − ( div ω A ) ω \partial A' = \partial(A\mathbin{\lrcorner}\omega) = -(\operatorname{div}_{\omega} A)\mathbin{\lrcorner}\omega= -(\operatorname{div}_{\omega} A)\omega ∂ A ′ = ∂ ( A ┘ ω ) = − ( div ω A ) ┘ ω = − ( div ω A ) ω , where in the last step we used that div ω A ∈ P V 0 , 1 \operatorname{div}_{\omega} A \in\mathrm{PV}^{0,1} div ω A ∈ PV 0 , 1 , and contraction of a scalar-valued form with ω \omega ω is multiplication. This constraint reduces the gauge transformations to ‘volume-preserving’ diffeomorphisms with the vector parameter ξ i \xi^i ξ i being divergence free, on which the dg Lie bracket closes.
We then use that for ∂ A ′ = 0 \partial A' = 0 ∂ A ′ = 0 it follows that ∂ ˉ A ′ \bar{\partial} A' ∂ ˉ A ′ is ∂ \partial ∂ -exact. Indeed, the ( 2 , 2 ) (2,2) ( 2 , 2 ) -form ∂ ˉ A ′ \bar{\partial} A' ∂ ˉ A ′ is ∂ ˉ \bar{\partial} ∂ ˉ -closed and ∂ ˉ \bar{\partial} ∂ ˉ -exact, and it is ∂ \partial ∂ -closed since ∂ ∂ ˉ A ′ = − ∂ ˉ ∂ A ′ = 0 \partial\bar{\partial} A' = -\bar{\partial}\partial A' = 0 ∂ ∂ ˉ A ′ = − ∂ ˉ ∂ A ′ = 0 . By the ∂ ∂ ˉ \partial\bar{\partial} ∂ ∂ ˉ -lemma of § 2 it is therefore of the form ∂ ˉ A ′ = ∂ ∂ ˉ γ \bar{\partial} A' = \partial\bar{\partial}\gamma ∂ ˉ A ′ = ∂ ∂ ˉ γ for some γ ∈ Ω 1 , 1 \gamma\in\Omega^{1,1} γ ∈ Ω 1 , 1 , both on a compact Calabi–Yau threefold and locally. Setting Ψ = ∂ ˉ γ ∈ Ω 1 , 2 \Psi= \bar{\partial}\gamma\in\Omega^{1,2} Ψ = ∂ ˉ γ ∈ Ω 1 , 2 we thus have ∂ ˉ A ′ = ∂ Ψ \bar{\partial} A' = \partial\Psi ∂ ˉ A ′ = ∂ Ψ , and we define
∂ − 1 ∂ ˉ A ′ : = Ψ . (31) \partial^{-1}\bar{\partial} A' := \Psi. \tag*{(31)} ∂ − 1 ∂ ˉ A ′ := Ψ. (31)
The form γ \gamma γ is determined up to γ 0 \gamma_0 γ 0 with ∂ ∂ ˉ γ 0 = 0 \partial\bar{\partial}\gamma_0 = 0 ∂ ∂ ˉ γ 0 = 0 , which changes Ψ \Psi Ψ by ∂ ˉ γ 0 \bar{\partial}\gamma_0 ∂ ˉ γ 0 . The BCOV action for Kodaira–Spencer gravity for ∂ A ′ = 0 \partial A' = 0 ∂ A ′ = 0 can then be written as [3 ]
S [ A ′ ] = 1 2 ∫ X A ′ ∧ ∂ − 1 ∂ ˉ A ′ + 1 6 ∫ X ω ∧ ( ( A ∧ A ∧ A ) ⌟ ω ) , A = A ′ ⌟ ω − 1 . (32) S[A'] = \frac{1}{2}\int_X A' \wedge\partial^{-1}\bar{\partial} A' + \frac{1}{6}\int_X \omega\wedge((A\wedge A\wedge A)\mathbin{\lrcorner}\omega), \qquad A = A'\mathbin{\lrcorner}\omega^{-1} . \tag*{(32)} S [ A ′ ] = 2 1 ∫ X A ′ ∧ ∂ − 1 ∂ ˉ A ′ + 6 1 ∫ X ω ∧ (( A ∧ A ∧ A ) ┘ ω ) , A = A ′ ┘ ω − 1 . (32)
The quadratic term does not depend on the choice of γ \gamma γ . Indeed, integrating by parts on the compact manifold X X X , the change ∫ X A ′ ∧ ∂ ˉ γ 0 \int_X A'\wedge\bar{\partial}\gamma_0 ∫ X A ′ ∧ ∂ ˉ γ 0 equals, up to sign, ∫ X ∂ ˉ A ′ ∧ γ 0 = ∫ X ∂ ∂ ˉ γ ∧ γ 0 \int_X\bar{\partial}A'\wedge\gamma_0 = \int_X\partial\bar{\partial}\gamma\wedge\gamma_0 ∫ X ∂ ˉ A ′ ∧ γ 0 = ∫ X ∂ ∂ ˉ γ ∧ γ 0 , which after two further integrations by parts becomes ∫ X γ ∧ ∂ ∂ ˉ γ 0 = 0 \int_X\gamma\wedge\partial\bar{\partial}\gamma_0 = 0 ∫ X γ ∧ ∂ ∂ ˉ γ 0 = 0 . Variation of the action yields
δ S = ∫ X δ A ′ ∧ ( ∂ − 1 ∂ ˉ A ′ − 1 2 ( A ∧ A ) ⌟ ω ) , (33) \delta S = \int_X \delta A' \wedge\left(\partial^{-1}\bar{\partial} A' - \frac{1}{2}(A\wedge A)\mathbin{\lrcorner}\omega\right), \tag*{(33)} δ S = ∫ X δ A ′ ∧ ( ∂ − 1 ∂ ˉ A ′ − 2 1 ( A ∧ A ) ┘ ω ) , (33)
and acting on the resulting equations of motion with ∂ \partial ∂ we have
∂ ˉ A ′ − 1 2 ∂ ( ( A ∧ A ) ⌟ ω ) = 0. (34) \bar{\partial} A' - \frac{1}{2}\partial((A\wedge A)\mathbin{\lrcorner}\omega) = 0 . \tag*{(34)} ∂ ˉ A ′ − 2 1 ∂ (( A ∧ A ) ┘ ω ) = 0. (34)
Using (20) and div ω A = 0 \operatorname{div}_{\omega} A = 0 div ω A = 0 it follows that ∂ ( ( A ∧ A ) ⌟ ω ) = − { A , A } \partial((A\wedge A)\mathbin{\lrcorner}\omega) = -\{A,A\} ∂ (( A ∧ A ) ┘ ω ) = − { A , A } , and so the above equations are equivalent to the Kodaira–Spencer equations (27) in the form ( ∂ ˉ A + 1 2 { A , A } ) ⌟ ω = 0 (\bar{\partial}A + \frac{1}{2}\{A,A\})\mathbin{\lrcorner}\omega= 0 ( ∂ ˉ A + 2 1 { A , A }) ┘ ω = 0 .
Local action for Kodaira–Spencer gravity
In this section we present the local action principle for Kodaira–Spencer gravity formulated as an L ∞ L_{\infty} L ∞ algebra. In the first subsection, we introduce the Leibniz–Loday algebra of holomorphic one-forms, which closely follows [27 ] and defines a Courant algebroid and an associated L ∞ L_{\infty} L ∞ algebra due to Roytenberg and Weinstein. In the second subsection, we present an L ∞ L_{\infty} L ∞ algebra on a subcomplex of the Dolbeault bicomplex which encodes the field equations, the gauge symmetries and their algebra. In the third subsection, we relate this L ∞ L_{\infty} L ∞ algebra to a cyclic one and give the corresponding BV master action. In the fourth subsection, we reduce the theory to three dimensions. Throughout this section we ignore global issues, as appropriate for instance for X = C 3 X = \mathbb{C}^3 X = C 3 , leaving the inclusion of cohomology and complex structure moduli to § 4.3 .
Leibniz–Loday algebra and associated L structure
As outlined in the introduction we now introduce a new ‘prepotential’ field by solving (30) :
A ′ = ∂ e ⟺ A j ˉ i = ω i j k ∂ j e k j ˉ , (35) A' = \partial e \qquad\Longleftrightarrow\qquad A_{\bar{j}}^{i} = \omega^{ijk}\partial_j e_{k\bar{j}}, \tag*{(35)} A ′ = ∂ e ⟺ A j ˉ i = ω ij k ∂ j e k j ˉ , (35)
which in turn introduces a new gauge redundancy in addition to that encoded in (28) . The corresponding gauge algebra is governed by a Leibniz–Loday algebra originating from an embedding tensor map ϑ : Ω 1 , 0 → Γ ( T 1 , 0 ) \vartheta: \Omega^{1,0} \to\Gamma(T^{1,0}) ϑ : Ω 1 , 0 → Γ ( T 1 , 0 ) from holomorphic one-forms to vectors, defined by
( ϑ ( α ) ) i = ω i j k ∂ j α k ⟺ ι ϑ ( α ) ω = ∂ α . (36) \bigl(\vartheta(\alpha)\bigr)^i = \omega^{ijk}\partial_j\alpha_k \qquad\Longleftrightarrow\qquad\iota_{\vartheta(\alpha)}\omega= \partial\alpha. \tag*{(36)} ( ϑ ( α ) ) i = ω ij k ∂ j α k ⟺ ι ϑ ( α ) ω = ∂ α . (36)
These vector fields are divergence-free,
div ω ( ϑ ( α ) ) = 0. (37) \operatorname{div}_{\omega}(\vartheta(\alpha)) = 0 . \tag*{(37)} div ω ( ϑ ( α )) = 0. (37)
Furthermore, the embedding tensor map has a non-trivial kernel given by
ker ϑ = ∂ Ω 0 , 0 = { ∂ α = 0 } . (38) \ker\vartheta= \partial\Omega^{0,0} = \{\partial\alpha= 0\} . \tag*{(38)} ker ϑ = ∂ Ω 0 , 0 = { ∂ α = 0 } . (38)
Next, we define the Leibniz–Loday algebra on this space of ( 1 , 0 ) (1,0) ( 1 , 0 ) forms as the bilinear ‘product’
α 1 ∘ α 2 : = L ϑ ( α 1 ) α 2 , (39) \alpha_1 \circ\alpha_2 := \mathcal{L}_{\vartheta(\alpha_1)}\alpha_2 , \tag*{(39)} α 1 ∘ α 2 := L ϑ ( α 1 ) α 2 , (39)
where the Lie derivative acts on the holomorphic sector as L X = ι X ∂ + ∂ ι X \mathcal{L}_X = \iota_X\partial+ \partial\iota_X L X = ι X ∂ + ∂ ι X . Note that this operation is not antisymmetric. Rather, it has a symmetric part, which we will see momentarily is encoded in the following bilinear operation, written in the notation of [21 , 24 ] :
α 1 ∙ α 2 : = ι ϑ ( α 1 ) α 2 + ι ϑ ( α 2 ) α 1 ∈ Ω 0 , 0 . (40) \alpha_1 \bullet\alpha_2 := \iota_{\vartheta(\alpha_1)}\alpha_2 + \iota_{\vartheta(\alpha_2)}\alpha_1 \in\Omega^{0,0} . \tag*{(40)} α 1 ∙ α 2 := ι ϑ ( α 1 ) α 2 + ι ϑ ( α 2 ) α 1 ∈ Ω 0 , 0 . (40)
The properties of Leibniz–Loday algebras are encoded in the following series of propositions:
Proposition 1 (Quadratic constraint):
ϑ ( α 1 ∘ α 2 ) = [ ϑ ( α 1 ) , ϑ ( α 2 ) ] . (41) \vartheta(\alpha_1 \circ\alpha_2) = [\vartheta(\alpha_1),\vartheta(\alpha_2)] . \tag*{(41)} ϑ ( α 1 ∘ α 2 ) = [ ϑ ( α 1 ) , ϑ ( α 2 )] . (41) Proof. By (36) and ∂ ω = 0 \partial\omega= 0 ∂ ω = 0 we have L ϑ ( α 1 ) ω = ι ϑ ( α 1 ) ∂ ω + ∂ ( ι ϑ ( α 1 ) ω ) = ∂ ∂ α 1 = 0 \mathcal{L}_{\vartheta(\alpha_1)}\omega= \iota_{\vartheta(\alpha_1)}\partial\omega+ \partial(\iota_{\vartheta(\alpha_1)}\omega) = \partial\partial\alpha_1 = 0 L ϑ ( α 1 ) ω = ι ϑ ( α 1 ) ∂ ω + ∂ ( ι ϑ ( α 1 ) ω ) = ∂∂ α 1 = 0 . Using ∂ α 2 = ι ϑ ( α 2 ) ω \partial\alpha_2 = \iota_{\vartheta(\alpha_2)}\omega ∂ α 2 = ι ϑ ( α 2 ) ω it follows that L ϑ ( α 1 ) ( ∂ α 2 ) = L ϑ ( α 1 ) ι ϑ ( α 2 ) ω = ι [ ϑ ( α 1 ) , ϑ ( α 2 ) ] ω + ι ϑ ( α 2 ) L ϑ ( α 1 ) ω = ι [ ϑ ( α 1 ) , ϑ ( α 2 ) ] ω \mathcal{L}_{\vartheta(\alpha_1)}(\partial\alpha_2) = \mathcal{L}_{\vartheta(\alpha_1)}\iota_{\vartheta(\alpha_2)}\omega= \iota_{[\vartheta(\alpha_1),\vartheta(\alpha_2)]}\omega+ \iota_{\vartheta(\alpha_2)}\mathcal{L}_{\vartheta(\alpha_1)}\omega= \iota_{[\vartheta(\alpha_1),\vartheta(\alpha_2)]}\omega L ϑ ( α 1 ) ( ∂ α 2 ) = L ϑ ( α 1 ) ι ϑ ( α 2 ) ω = ι [ ϑ ( α 1 ) , ϑ ( α 2 )] ω + ι ϑ ( α 2 ) L ϑ ( α 1 ) ω = ι [ ϑ ( α 1 ) , ϑ ( α 2 )] ω . Using [ ∂ , L X ] = 0 [\partial,\mathcal{L}_X] = 0 [ ∂ , L X ] = 0 and (36) , we have ι ϑ ( α 1 ∘ α 2 ) ω = ∂ ( α 1 ∘ α 2 ) = L ϑ ( α 1 ) ∂ α 2 = ι [ ϑ ( α 1 ) , ϑ ( α 2 ) ] ω \iota_{\vartheta(\alpha_1\circ\alpha_2)}\omega= \partial(\alpha_1 \circ\alpha_2) = \mathcal{L}_{\vartheta(\alpha_1)}\partial\alpha_2 = \iota_{[\vartheta(\alpha_1),\vartheta(\alpha_2)]}\omega ι ϑ ( α 1 ∘ α 2 ) ω = ∂ ( α 1 ∘ α 2 ) = L ϑ ( α 1 ) ∂ α 2 = ι [ ϑ ( α 1 ) , ϑ ( α 2 )] ω . The claim then follows with nondegeneracy of ω \omega ω .
Proposition 2 (Leibniz–Loday identity):
α 1 ∘ ( α 2 ∘ α 3 ) = ( α 1 ∘ α 2 ) ∘ α 3 + α 2 ∘ ( α 1 ∘ α 3 ) . (42) \alpha_1 \circ(\alpha_2 \circ\alpha_3) = (\alpha_1 \circ\alpha_2) \circ\alpha_3 + \alpha_2 \circ(\alpha_1 \circ\alpha_3) . \tag*{(42)} α 1 ∘ ( α 2 ∘ α 3 ) = ( α 1 ∘ α 2 ) ∘ α 3 + α 2 ∘ ( α 1 ∘ α 3 ) . (42) Proof. The left-hand side is L ϑ ( α 1 ) L ϑ ( α 2 ) α 3 \mathcal{L}_{\vartheta(\alpha_1)}\mathcal{L}_{\vartheta(\alpha_2)}\alpha_3 L ϑ ( α 1 ) L ϑ ( α 2 ) α 3 . By Proposition 1, ( α 1 ∘ α 2 ) ∘ α 3 = L [ ϑ ( α 1 ) , ϑ ( α 2 ) ] α 3 (\alpha_1 \circ\alpha_2) \circ\alpha_3 = \mathcal{L}_{[\vartheta(\alpha_1),\vartheta(\alpha_2)]}\alpha_3 ( α 1 ∘ α 2 ) ∘ α 3 = L [ ϑ ( α 1 ) , ϑ ( α 2 )] α 3 , so the right-hand side is L [ ϑ ( α 1 ) , ϑ ( α 2 ) ] α 3 + L ϑ ( α 2 ) L ϑ ( α 1 ) α 3 = L ϑ ( α 1 ) L ϑ ( α 2 ) α 3 \mathcal{L}_{[\vartheta(\alpha_1),\vartheta(\alpha_2)]}\alpha_3 + \mathcal{L}_{\vartheta(\alpha_2)}\mathcal{L}_{\vartheta(\alpha_1)}\alpha_3 = \mathcal{L}_{\vartheta(\alpha_1)}\mathcal{L}_{\vartheta(\alpha_2)}\alpha_3 L [ ϑ ( α 1 ) , ϑ ( α 2 )] α 3 + L ϑ ( α 2 ) L ϑ ( α 1 ) α 3 = L ϑ ( α 1 ) L ϑ ( α 2 ) α 3 using L [ ϑ ( α 1 ) , ϑ ( α 2 ) ] = [ L ϑ ( α 1 ) , L ϑ ( α 2 ) ] \mathcal{L}_{[\vartheta(\alpha_1),\vartheta(\alpha_2)]} = [\mathcal{L}_{\vartheta(\alpha_1)},\mathcal{L}_{\vartheta(\alpha_2)}] L [ ϑ ( α 1 ) , ϑ ( α 2 )] = [ L ϑ ( α 1 ) , L ϑ ( α 2 ) ] .
Proposition 3 (Symmetric part is trivial):
α 1 ∘ α 2 + α 2 ∘ α 1 = ∂ ( α 1 ∙ α 2 ) ∈ ∂ Ω 0 , 0 = ker ϑ . (43) \alpha_1 \circ\alpha_2 + \alpha_2 \circ\alpha_1 = \partial(\alpha_1 \mathbin{\bullet} \alpha_2) \in\partial\Omega^{0,0} = \ker\vartheta. \tag*{(43)} α 1 ∘ α 2 + α 2 ∘ α 1 = ∂ ( α 1 ∙ α 2 ) ∈ ∂ Ω 0 , 0 = ker ϑ . (43)
Proof: L ϑ ( α 1 ) α 2 + L ϑ ( α 2 ) α 1 = ι ϑ ( α 1 ) ∂ α 2 + ι ϑ ( α 2 ) ∂ α 1 + ∂ ( ι ϑ ( α 1 ) α 2 + ι ϑ ( α 2 ) α 1 ) \mathcal{L}_{\vartheta(\alpha_1)}\alpha_2+\mathcal{L}_{\vartheta(\alpha_2)}\alpha_1=\iota_{\vartheta(\alpha_1)}\partial\alpha_2+\iota_{\vartheta(\alpha_2)}\partial\alpha_1+\partial(\iota_{\vartheta(\alpha_1)}\alpha_2+\iota_{\vartheta(\alpha_2)}\alpha_1) L ϑ ( α 1 ) α 2 + L ϑ ( α 2 ) α 1 = ι ϑ ( α 1 ) ∂ α 2 + ι ϑ ( α 2 ) ∂ α 1 + ∂ ( ι ϑ ( α 1 ) α 2 + ι ϑ ( α 2 ) α 1 ) . The first two cancel: ι ϑ ( α 1 ) ι ϑ ( α 2 ) ω + ι ϑ ( α 2 ) ι ϑ ( α 1 ) ω = 0 \iota_{\vartheta(\alpha_1)}\iota_{\vartheta(\alpha_2)}\omega+\iota_{\vartheta(\alpha_2)}\iota_{\vartheta(\alpha_1)}\omega=0 ι ϑ ( α 1 ) ι ϑ ( α 2 ) ω + ι ϑ ( α 2 ) ι ϑ ( α 1 ) ω = 0 . The remainder is ∂ ( α 1 ∙ α 2 ) \partial(\alpha_1\mathbin{\bullet}\alpha_2) ∂ ( α 1 ∙ α 2 ) by (39) . □
We have now established that the above structure exhibits a Courant algebroid structure in the sense of [28 ] .3 Indeed, with ϑ ∘ ∂ = 0 \vartheta\circ\partial=0 ϑ ∘ ∂ = 0 , cf. eq. (38) , the Courant axiom 4 of [28 Def. 3.2] and ∂ f ∙ α = ρ ( α ) f \partial f\mathbin{\bullet}\alpha=\rho(\alpha)f ∂ f ∙ α = ρ ( α ) f , where here ρ = ϑ \rho=\vartheta ρ = ϑ , Propositions 1–3 exhibit ( Ω 1 , 0 ; ϑ , ∙ , [ ⋅ , ⋅ ] C ) (\Omega^{1,0};\vartheta,\mathbin{\bullet},[\cdot,\cdot]_C) ( Ω 1 , 0 ; ϑ , ∙ , [ ⋅ , ⋅ ] C ) as the holomorphic Courant algebroid [27 ] , with Courant bracket
[ α 1 , α 2 ] C ≔ 1 2 ( α 1 ∘ α 2 − α 2 ∘ α 1 ) . (44) [\alpha_1,\alpha_2]_C\coloneqq\frac{1}{2}(\alpha_1\circ\alpha_2-\alpha_2\circ\alpha_1). \tag*{(44)} [ α 1 , α 2 ] C : = 2 1 ( α 1 ∘ α 2 − α 2 ∘ α 1 ) . (44)
By [28 Thm. 4.3] a Courant algebroid yields an L ∞ L_\infty L ∞ algebra. Here this L ∞ L_\infty L ∞ algebra lives on the two-term complex
L R W ≡ ( Ω 0 , 0 → ∂ Ω 1 , 0 ) , (45) L_{\mathrm{RW}}\equiv\left(\Omega^{0,0}\xrightarrow{\partial}\Omega^{1,0}\right), \tag*{(45)} L RW ≡ ( Ω 0 , 0 ∂ Ω 1 , 0 ) , (45)
where the degrees are deg Ω 0 , 0 = − 1 \deg\Omega^{0,0}=-1 deg Ω 0 , 0 = − 1 , deg Ω 1 , 0 = 0 \deg\Omega^{1,0}=0 deg Ω 1 , 0 = 0 . The L ∞ L_\infty L ∞ structure maps are given by ℓ 1 = ∂ \ell_1=\partial ℓ 1 = ∂ , together with
ℓ 2 ( α 1 , α 2 ) = [ α 1 , α 2 ] C , \ell_2(\alpha_1,\alpha_2)=[\alpha_1,\alpha_2]_C, ℓ 2 ( α 1 , α 2 ) = [ α 1 , α 2 ] C ,
ℓ 2 ( α , f ) = 1 2 α ∙ ∂ f = 1 2 ι ϑ ( α ) ∂ f , \ell_2(\alpha,f)=\frac{1}{2}\alpha\mathbin{\bullet}\partial f=\frac{1}{2}\iota_{\vartheta(\alpha)}\partial f, ℓ 2 ( α , f ) = 2 1 α ∙ ∂ f = 2 1 ι ϑ ( α ) ∂ f ,
ℓ 3 ( α 1 , α 2 , α 3 ) = − T ( α 1 , α 2 , α 3 ) , (46) \ell_3(\alpha_1,\alpha_2,\alpha_3)=-T(\alpha_1,\alpha_2,\alpha_3), \tag*{(46)} ℓ 3 ( α 1 , α 2 , α 3 ) = − T ( α 1 , α 2 , α 3 ) , (46)
and Nijenhuis function [28 ]
T = 1 2 [ α [ 1 , α 2 ] ] C ∙ α 3 = 1 6 ( [ α 1 , α 2 ] C ∙ α 3 + two more ) , (47) T=\frac{1}{2}[\alpha_{[1},\alpha_{2]}]_C\mathbin{\bullet}\alpha_3=\frac{1}{6}([\alpha_1,\alpha_2]_C\mathbin{\bullet}\alpha_3+\text{two more}), \tag*{(47)} T = 2 1 [ α [ 1 , α 2 ] ] C ∙ α 3 = 6 1 ([ α 1 , α 2 ] C ∙ α 3 + two more ) , (47)
using antisymmetrizations with combinatorial weight factors ( A [ 1 B 2 ] = 1 2 ( A 1 B 2 − A 2 B 1 ) ) (A_{[1}B_{2]}=\frac{1}{2}(A_1B_2-A_2B_1)) ( A [ 1 B 2 ] = 2 1 ( A 1 B 2 − A 2 B 1 )) , etc.). All ℓ ≥ 4 \ell_{\geq4} ℓ ≥ 4 are zero.4
The L ∞ L_\infty L ∞ relations follow by direct computation. For instance, the Leibniz relation requires by degree reasons
ℓ 1 ( ℓ 2 ( α , f ) ) = ℓ 2 ( ℓ 1 ( α ) , f ) + ( − 1 ) ∣ α ∣ ℓ 2 ( α , ℓ 1 ( f ) ) = ℓ 2 ( α , ℓ 1 ( f ) ) . (48) \ell_1(\ell_2(\alpha,f))=\ell_2(\ell_1(\alpha),f)+(-1)^{|\alpha|}\ell_2(\alpha,\ell_1(f))=\ell_2(\alpha,\ell_1(f)). \tag*{(48)} ℓ 1 ( ℓ 2 ( α , f )) = ℓ 2 ( ℓ 1 ( α ) , f ) + ( − 1 ) ∣ α ∣ ℓ 2 ( α , ℓ 1 ( f )) = ℓ 2 ( α , ℓ 1 ( f )) . (48)
For the left-hand side we compute
ℓ 1 ( ℓ 2 ( α , f ) ) = ∂ ( ℓ 2 ( α , f ) ) = 1 2 ∂ ι ϑ ( α ) ∂ f , (49) \ell_1(\ell_2(\alpha,f))=\partial(\ell_2(\alpha,f))=\frac{1}{2}\partial\iota_{\vartheta(\alpha)}\partial f, \tag*{(49)} ℓ 1 ( ℓ 2 ( α , f )) = ∂ ( ℓ 2 ( α , f )) = 2 1 ∂ ι ϑ ( α ) ∂ f , (49)
and for the right-hand side
ℓ 2 ( α , ℓ 1 ( f ) ) = [ α , ∂ f ] C = 1 2 ( α ∘ ∂ f − ∂ f ∘ α ) = 1 2 α ∘ ∂ f = 1 2 L ϑ ( α ) ∂ f = 1 2 ι ϑ ( α ) ∂ ∂ f + 1 2 ∂ ι ϑ ( α ) ∂ f = 1 2 ∂ ι ϑ ( α ) ∂ f , (50) \begin{aligned}
\ell_2(\alpha,\ell_1(f))=[\alpha,\partial f]_C=\frac{1}{2}(\alpha\circ\partial f-\partial f\circ\alpha)=\frac{1}{2}\alpha\circ\partial f=\frac{1}{2}\mathcal{L}_{\vartheta(\alpha)}\partial f \\
&=\frac{1}{2}\iota_{\vartheta(\alpha)}\partial\partial f+\frac{1}{2}\partial\iota_{\vartheta(\alpha)}\partial f=\frac{1}{2}\partial\iota_{\vartheta(\alpha)}\partial f, \tag*{(50)}
\end{aligned} ℓ 2 ( α , ℓ 1 ( f )) = [ α , ∂ f ] C = 2 1 ( α ∘ ∂ f − ∂ f ∘ α ) = 2 1 α ∘ ∂ f = 2 1 L ϑ ( α ) ∂ f = 2 1 ι ϑ ( α ) ∂∂ f + 2 1 ∂ ι ϑ ( α ) ∂ f = 2 1 ∂ ι ϑ ( α ) ∂ f , (50)
which equals (3.15).
L ∞ L_\infty L ∞ algebra on the bicomplex and the local action
We now present the L ∞ L_\infty L ∞ algebra that encodes the field equations, gauge symmetries and gauge algebra of the local theory. Let us first recall the relevant definitions [30 ] . An L ∞ L_\infty L ∞ algebra is a graded vector space L = ⨁ k L k L=\bigoplus_k L_k L = ⨁ k L k with multilinear maps ℓ n : L ⊗ n → L \ell_n:L^{\otimes n}\to L ℓ n : L ⊗ n → L of degree 2 − n 2-n 2 − n that are graded antisymmetric,
ℓ n ( … , x i , x i + 1 , … ) = − ( − 1 ) x i x i + 1 ℓ n ( … , x i + 1 , x i , … ) , (51) \ell_n(\ldots,x_i,x_{i+1},\ldots)=-(-1)^{x_i x_{i+1}}\ell_n(\ldots,x_{i+1},x_i,\ldots), \tag*{(51)} ℓ n ( … , x i , x i + 1 , … ) = − ( − 1 ) x i x i + 1 ℓ n ( … , x i + 1 , x i , … ) , (51)
where we recall that in sign exponents x x x denotes the degree ∣ x ∣ \lvert x\rvert ∣ x ∣ . The maps are subject to the L ∞ L_\infty L ∞ relations
∑ i + j = n + 1 ( − 1 ) i ( j − 1 ) ∑ σ ( − 1 ) σ ϵ ( σ ; x ) ℓ j ( ℓ i ( x σ ( 1 ) , … , x σ ( i ) ) , x σ ( i + 1 ) , … , x σ ( n ) ) = 0 , (52) \sum_{i+j=n+1}(-1)^{i(j-1)}\sum_{\sigma}(-1)^\sigma\epsilon(\sigma;x)\,\ell_j\bigl(\ell_i(x_{\sigma(1)},\ldots,x_{\sigma(i)}),x_{\sigma(i+1)},\ldots,x_{\sigma(n)}\bigr)=0, \tag*{(52)} i + j = n + 1 ∑ ( − 1 ) i ( j − 1 ) σ ∑ ( − 1 ) σ ϵ ( σ ; x ) ℓ j ( ℓ i ( x σ ( 1 ) , … , x σ ( i ) ) , x σ ( i + 1 ) , … , x σ ( n ) ) = 0 , (52)
with the inner sum running over ( i , n − i ) (i,n-i) ( i , n − i ) -unshuffles, where ϵ ( σ ; x ) \epsilon(\sigma;x) ϵ ( σ ; x ) is the Koszul sign of the permutation. The first three relations state that ℓ 1 \ell_1 ℓ 1 squares to zero, that ℓ 1 \ell_1 ℓ 1 is a derivation of ℓ 2 \ell_2 ℓ 2 ,
ℓ 1 ℓ 2 ( x 1 , x 2 ) = ℓ 2 ( ℓ 1 x 1 , x 2 ) + ( − 1 ) x 1 ℓ 2 ( x 1 , ℓ 1 x 2 ) , (53) \ell_1\ell_2(x_1,x_2)=\ell_2(\ell_1x_1,x_2)+(-1)^{x_1}\ell_2(x_1,\ell_1x_2), \tag*{(53)} ℓ 1 ℓ 2 ( x 1 , x 2 ) = ℓ 2 ( ℓ 1 x 1 , x 2 ) + ( − 1 ) x 1 ℓ 2 ( x 1 , ℓ 1 x 2 ) , (53)
and that the Jacobi identity of ℓ 2 \ell_2 ℓ 2 holds up to ℓ 1 \ell_1 ℓ 1 -exact terms governed by ℓ 3 \ell_3 ℓ 3 ,
ℓ 2 ( ℓ 2 ( x 1 , x 2 ) , x 3 ) + ( − 1 ) ( x 1 + x 2 ) x 3 ℓ 2 ( ℓ 2 ( x 3 , x 1 ) , x 2 ) + ( − 1 ) ( x 2 + x 3 ) x 1 ℓ 2 ( ℓ 2 ( x 2 , x 3 ) , x 1 ) = − ℓ 1 ℓ 3 ( x 1 , x 2 , x 3 ) − ℓ 3 ( ℓ 1 x 1 , x 2 , x 3 ) − ( − 1 ) x 1 ℓ 3 ( x 1 , ℓ 1 x 2 , x 3 ) − ( − 1 ) x 1 + x 2 ℓ 3 ( x 1 , x 2 , ℓ 1 x 3 ) . (54) \begin{aligned}
\ell_2(\ell_2(x_1,x_2),x_3)&+(-1)^{(x_1+x_2)x_3}\ell_2(\ell_2(x_3,x_1),x_2)+(-1)^{(x_2+x_3)x_1}\ell_2(\ell_2(x_2,x_3),x_1)\\
&=-\ell_1\ell_3(x_1,x_2,x_3)-\ell_3(\ell_1x_1,x_2,x_3)\\
&\quad-(-1)^{x_1}\ell_3(x_1,\ell_1x_2,x_3)-(-1)^{x_1+x_2}\ell_3(x_1,x_2,\ell_1x_3). \tag*{(54)}
\end{aligned} ℓ 2 ( ℓ 2 ( x 1 , x 2 ) , x 3 ) + ( − 1 ) ( x 1 + x 2 ) x 3 ℓ 2 ( ℓ 2 ( x 3 , x 1 ) , x 2 ) + ( − 1 ) ( x 2 + x 3 ) x 1 ℓ 2 ( ℓ 2 ( x 2 , x 3 ) , x 1 ) = − ℓ 1 ℓ 3 ( x 1 , x 2 , x 3 ) − ℓ 3 ( ℓ 1 x 1 , x 2 , x 3 ) − ( − 1 ) x 1 ℓ 3 ( x 1 , ℓ 1 x 2 , x 3 ) − ( − 1 ) x 1 + x 2 ℓ 3 ( x 1 , x 2 , ℓ 1 x 3 ) . (54)
The L ∞ L_\infty L ∞ algebra of the local theory lives on the following eight spaces of the Dolbeault bicomplex, which are exchanged in pairs by Ω p , q ↔ Ω 3 − p , 3 − q \Omega^{p,q}\leftrightarrow\Omega^{3-p,3-q} Ω p , q ↔ Ω 3 − p , 3 − q :
Figure 1.
The content is the gauge-for-gauge parameter χ ∈ Ω 0 , 0 \chi\in\Omega^{0,0} χ ∈ Ω 0 , 0 , the gauge parameters λ ∈ Ω 1 , 0 \lambda\in\Omega^{1,0} λ ∈ Ω 1 , 0 (diffeomorphisms) and λ ˉ ∈ Ω 0 , 1 \bar{\lambda}\in\Omega^{0,1} λ ˉ ∈ Ω 0 , 1 (the redundancy of the prepotential), the field e ∈ Ω 1 , 1 e\in\Omega^{1,1} e ∈ Ω 1 , 1 , and the spaces of ‘antifields’ e ∗ ∈ Ω 2 , 2 e^*\in\Omega^{2,2} e ∗ ∈ Ω 2 , 2 , λ ∗ ∈ Ω 2 , 3 \lambda^*\in\Omega^{2,3} λ ∗ ∈ Ω 2 , 3 , λ ˉ ∗ ∈ Ω 3 , 2 \bar{\lambda}^*\in\Omega^{3,2} λ ˉ ∗ ∈ Ω 3 , 2 and χ ∗ ∈ Ω 3 , 3 \chi^*\in\Omega^{3,3} χ ∗ ∈ Ω 3 , 3 . The antifield spaces can be interpreted as the spaces of field equations (e ∗ e^* e ∗ ), of Noether identities (λ ∗ \lambda^* λ ∗ and λ ˉ ∗ \bar{\lambda}^* λ ˉ ∗ , one for each gauge parameter) and of Noether-for-Noether identities (χ ∗ \chi^* χ ∗ ). Their degrees are − 1 -1 − 1 for χ \chi χ , 0 0 0 for λ \lambda λ and λ ˉ \bar{\lambda} λ ˉ , 1 1 1 for e e e , 2 2 2 for e ∗ e^* e ∗ , 3 3 3 for λ ∗ \lambda^* λ ∗ and λ ˉ ∗ \bar{\lambda}^* λ ˉ ∗ , and 4 4 4 for χ ∗ \chi^* χ ∗ . Equivalently, for each space Ω p , q \Omega^{p,q} Ω p , q the degree is p + q − 1 p+q-1 p + q − 1 for the four spaces on the left and p + q − 2 p+q-2 p + q − 2 for the four spaces on the right. The differential is
ℓ 1 ( χ ) = ( ∂ ˉ χ ∂ χ ) , ℓ 1 ( λ ˉ λ ) = ∂ λ ˉ + ∂ ˉ λ , ℓ 1 ( e ) = ∂ ∂ ˉ e , (55) \ell_1(\chi)=\begin{pmatrix}\bar{\partial}\chi\\ \partial\chi\end{pmatrix},\qquad
\ell_1\begin{pmatrix}\bar{\lambda}\\ \lambda\end{pmatrix}=\partial\bar{\lambda}+\bar{\partial}\lambda,\qquad
\ell_1(e)=\partial\bar{\partial}e, \tag*{(55)} ℓ 1 ( χ ) = ( ∂ ˉ χ ∂ χ ) , ℓ 1 ( λ ˉ λ ) = ∂ λ ˉ + ∂ ˉ λ , ℓ 1 ( e ) = ∂ ∂ ˉ e , (55)
ℓ 1 ( e ∗ ) = ( ∂ ˉ e ∗ ∂ e ∗ ) , ℓ 1 ( λ ∗ λ ˉ ∗ ) = ∂ λ ∗ + ∂ ˉ λ ˉ ∗ , ℓ 1 ( χ ∗ ) = 0 , (56) \ell_1(e^*)=\begin{pmatrix}\bar{\partial}e^*\\ \partial e^*\end{pmatrix},\qquad
\ell_1\begin{pmatrix}\lambda^*\\ \bar{\lambda}^*\end{pmatrix}=\partial\lambda^*+\bar{\partial}\bar{\lambda}^*,\qquad
\ell_1(\chi^*)=0, \tag*{(56)} ℓ 1 ( e ∗ ) = ( ∂ ˉ e ∗ ∂ e ∗ ) , ℓ 1 ( λ ∗ λ ˉ ∗ ) = ∂ λ ∗ + ∂ ˉ λ ˉ ∗ , ℓ 1 ( χ ∗ ) = 0 , (56)
where, as in the figure, the upper entries are λ ˉ ∈ Ω 0 , 1 \bar{\lambda}\in\Omega^{0,1} λ ˉ ∈ Ω 0 , 1 and λ ∗ ∈ Ω 2 , 3 \lambda^*\in\Omega^{2,3} λ ∗ ∈ Ω 2 , 3 and the lower entries λ ∈ Ω 1 , 0 \lambda\in\Omega^{1,0} λ ∈ Ω 1 , 0 and λ ˉ ∗ ∈ Ω 3 , 2 \bar{\lambda}^*\in\Omega^{3,2} λ ˉ ∗ ∈ Ω 3 , 2 . It squares to zero by (12) . It is first order except on the field, where it is the second-order operator ∂ ∂ ˉ \partial\bar{\partial} ∂ ∂ ˉ , thereby bridging the bidegrees ( 1 , 1 ) (1,1) ( 1 , 1 ) and ( 2 , 2 ) (2,2) ( 2 , 2 ) .
The higher brackets are as follows. We use the Schouten bracket (21) , the product (40) , and write c = λ ˉ ∗ ⌟ ω − 1 ∈ Ω 0 , 2 c=\bar{\lambda}^*\lrcorner\omega^{-1}\in\Omega^{0,2} c = λ ˉ ∗ ┘ ω − 1 ∈ Ω 0 , 2 for the antifield λ ˉ ∗ \bar{\lambda}^* λ ˉ ∗ with its holomorphic legs stripped off by ω − 1 \omega^{-1} ω − 1 . Operations of section 3.1 applied to e e e or λ ˉ \bar{\lambda} λ ˉ are understood to act on the holomorphic indices, at fixed antiholomorphic index. On the gauge parameters the binary bracket is the Courant bracket of section 3.1,
ℓ 2 ( λ 1 , λ 2 ) = [ λ 1 , λ 2 ] C , (57) \ell_2(\lambda_1,\lambda_2)=[\lambda_1,\lambda_2]_C, \tag*{(57)} ℓ 2 ( λ 1 , λ 2 ) = [ λ 1 , λ 2 ] C , (57)
and the remaining non-vanishing binary brackets are
ℓ 2 ( λ , χ ) = 1 2 ι ϑ ( λ ) ∂ χ , ℓ 2 ( λ , λ ˉ ) = 1 2 L ϑ ( λ ) λ ˉ , ℓ 2 ( χ , e ) = − 1 2 ι ϑ ( e ) ∂ χ , \ell_2(\lambda,\chi)=\frac{1}{2}\iota_{\vartheta(\lambda)}\partial\chi,\qquad
\ell_2(\lambda,\bar{\lambda})=\frac{1}{2}\mathcal{L}_{\vartheta(\lambda)}\bar{\lambda},\qquad
\ell_2(\chi,e)=-\frac{1}{2}\iota_{\vartheta(e)}\partial\chi, ℓ 2 ( λ , χ ) = 2 1 ι ϑ ( λ ) ∂ χ , ℓ 2 ( λ , λ ˉ ) = 2 1 L ϑ ( λ ) λ ˉ , ℓ 2 ( χ , e ) = − 2 1 ι ϑ ( e ) ∂ χ ,
ℓ 2 ( λ , e ) = [ λ , e ] C = L ϑ ( λ ) e − 1 2 ∂ ( λ ∙ e ) , \ell_2(\lambda,e)=[\lambda,e]_C=\mathcal{L}_{\vartheta(\lambda)}e-\frac{1}{2}\partial(\lambda\mathbin{\bullet}e), ℓ 2 ( λ , e ) = [ λ , e ] C = L ϑ ( λ ) e − 2 1 ∂ ( λ ∙ e ) ,
ℓ 2 ( λ , y ∗ ) = L ϑ ( λ ) y ∗ for y ∗ ∈ { e ∗ , λ ∗ , λ ˉ ∗ , χ ∗ } , \ell_2(\lambda,y^*)=\mathcal{L}_{\vartheta(\lambda)}y^*
\quad\text{for }y^*\in\{e^*,\lambda^*,\bar{\lambda}^*,\chi^*\}, ℓ 2 ( λ , y ∗ ) = L ϑ ( λ ) y ∗ for y ∗ ∈ { e ∗ , λ ∗ , λ ˉ ∗ , χ ∗ } ,
ℓ 2 ( e 1 , e 2 ) = − { ϑ ( e 1 ) , ϑ ( e 2 ) } ⌟ ω , ℓ 2 ( e , e ∗ ) = { ϑ ( e ) , e ∗ ⌟ ω − 1 } ⌟ ω , \ell_2(e_1,e_2)=-\{\vartheta(e_1),\vartheta(e_2)\}\lrcorner\omega,\qquad
\ell_2(e,e^*)=\{\vartheta(e),e^*\lrcorner\omega^{-1}\}\lrcorner\omega, ℓ 2 ( e 1 , e 2 ) = − { ϑ ( e 1 ) , ϑ ( e 2 )} ┘ ω , ℓ 2 ( e , e ∗ ) = { ϑ ( e ) , e ∗ ┘ ω − 1 } ┘ ω ,
ℓ 2 ( e , λ ˉ ∗ ) = ∂ ( ( ϑ ( e ) ∧ c ) ⌟ ω ) , (58) \ell_2(e,\bar{\lambda}^*)=\partial\bigl((\vartheta(e)\wedge c)\lrcorner\omega\bigr), \tag*{(58)} ℓ 2 ( e , λ ˉ ∗ ) = ∂ ( ( ϑ ( e ) ∧ c ) ┘ ω ) , (58)
where the Lie derivative acts on the holomorphic indices, and ι ϑ ( e ) ∂ χ \iota_{\vartheta(e)}\partial\chi ι ϑ ( e ) ∂ χ is the ( 0 , 1 ) (0,1) ( 0 , 1 ) -form obtained by contracting the vector-valued ( 0 , 1 ) (0,1) ( 0 , 1 ) -form ϑ ( e ) \vartheta(e) ϑ ( e ) with ∂ χ \partial\chi ∂ χ . The ternary brackets are those of the Roytenberg–Weinstein algebra,
ℓ 3 ( λ 1 , λ 2 , λ 3 ) = − T ( λ 1 , λ 2 , λ 3 ) , ℓ 3 ( λ 1 , λ 2 , e ) = − T ( λ 1 , λ 2 , e ) ∈ Ω 0 , 1 , (59) \ell_3(\lambda_1,\lambda_2,\lambda_3)=-T(\lambda_1,\lambda_2,\lambda_3),\qquad
\ell_3(\lambda_1,\lambda_2,e)=-T(\lambda_1,\lambda_2,e)\in\Omega^{0,1}, \tag*{(59)} ℓ 3 ( λ 1 , λ 2 , λ 3 ) = − T ( λ 1 , λ 2 , λ 3 ) , ℓ 3 ( λ 1 , λ 2 , e ) = − T ( λ 1 , λ 2 , e ) ∈ Ω 0 , 1 , (59)
and ℓ n ≥ 4 = 0 \ell_{n\geq4}=0 ℓ n ≥ 4 = 0 . All other brackets are either fixed by graded antisymmetry or vanish.
Proposition 4: The brackets (3.21)–(59) satisfy all L ∞ L_\infty L ∞ relations (52) .
We will call this L ∞ L_\infty L ∞ algebra L L L . Its relations follow from the double copy construction of section 4, in which they are inherited from those of its constituents. We have also verified them explicitly in components.
The field equation and the gauge transformations follow from the L ∞ L_\infty L ∞ algebra in the standard way. The field equation is the Maurer–Cartan equation5
E ( e ) : = ℓ 1 ( e ) + 1 2 ℓ 2 ( e , e ) = ∂ ∂ ˉ e − 1 2 { ϑ ( e ) , ϑ ( e ) } ⌟ ω = 0 , (60) E(e):=\ell_1(e)+\frac{1}{2}\ell_2(e,e)=\partial\bar{\partial}e-\frac{1}{2}\{\vartheta(e),\vartheta(e)\}\lrcorner\omega=0, \tag*{(60)} E ( e ) := ℓ 1 ( e ) + 2 1 ℓ 2 ( e , e ) = ∂ ∂ ˉ e − 2 1 { ϑ ( e ) , ϑ ( e )} ┘ ω = 0 , (60)
and it is covariant under the gauge transformations
δ e = ℓ 1 ( λ ˉ λ ) + ℓ 2 ( e , λ ) = ∂ ˉ λ + ∂ λ ˉ − [ λ , e ] C , (61) \delta e = \ell_1\begin{pmatrix}\bar{\lambda}\\ \lambda\end{pmatrix} + \ell_2(e,\lambda) = \bar{\partial}\lambda+ \partial\bar{\lambda} - [\lambda,e]_{\mathrm{C}}, \tag*{(61)} δ e = ℓ 1 ( λ ˉ λ ) + ℓ 2 ( e , λ ) = ∂ ˉ λ + ∂ λ ˉ − [ λ , e ] C , (61)
under which δ E = ℓ 2 ( E , λ ) \delta E = \ell_2(E,\lambda) δ E = ℓ 2 ( E , λ ) . Contracting with ω − 1 \omega^{-1} ω − 1 and using ( ∂ ∂ ˉ e ) ⌟ ω − 1 = − ∂ ˉ ( ϑ ( e ) ) (\partial\bar{\partial}e)\lrcorner\omega^{-1}=-\bar{\partial}(\vartheta(e)) ( ∂ ∂ ˉ e ) ┘ ω − 1 = − ∂ ˉ ( ϑ ( e )) , one finds E ( e ) ⌟ ω − 1 = − ( ∂ ˉ A + 1 2 { A , A } ) E(e)\lrcorner\omega^{-1}=-(\bar{\partial}A+\frac{1}{2}\{A,A\}) E ( e ) ┘ ω − 1 = − ( ∂ ˉ A + 2 1 { A , A }) with A = ϑ ( e ) A=\vartheta(e) A = ϑ ( e ) , so that (60) is precisely the Kodaira–Spencer equation (27) , written in terms of the prepotential. Similarly, since ϑ \vartheta ϑ annihilates the ∂ \partial ∂ -exact term in (58) , the gauge transformations (61) map under ϑ \vartheta ϑ to the gauge transformations (28) with parameter ξ = − ϑ ( λ ) \xi=-\vartheta(\lambda) ξ = − ϑ ( λ ) , i.e. to the volume-preserving diffeomorphisms of A A A , while the λ ˉ \bar{\lambda} λ ˉ -transformations are the redundancy of the prepotential. The algebra of gauge transformations closes off-shell, with the structure constants ℓ 2 ( λ 1 , λ 2 ) \ell_2(\lambda_1,\lambda_2) ℓ 2 ( λ 1 , λ 2 ) together with the field-dependent λ ˉ \bar{\lambda} λ ˉ -component ℓ 3 ( λ 1 , λ 2 , e ) \ell_3(\lambda_1,\lambda_2,e) ℓ 3 ( λ 1 , λ 2 , e ) . Moreover, the gauge algebra is reducible, with χ \chi χ parametrizing the trivial parameters ( λ , λ ˉ ) = ( ∂ χ , ∂ ˉ χ ˉ ) (\lambda,\bar{\lambda})=(\partial\chi,\bar{\partial}\bar{\chi}) ( λ , λ ˉ ) = ( ∂ χ , ∂ ˉ χ ˉ ) .
We finally note that the Courant bracket in (61) could be traded for the Lie derivative by the field-dependent parameter redefinition λ ˉ → λ ˉ + 1 2 λ ∙ e \bar{\lambda}\to\bar{\lambda}+\frac{1}{2}\lambda\bullet e λ ˉ → λ ˉ + 2 1 λ ∙ e , in terms of which δ e = ∂ ˉ λ + ∂ λ ˉ − L ϑ ( λ ) e \delta e=\bar{\partial}\lambda+\partial\bar{\lambda}-\mathcal{L}_{\vartheta(\lambda)}e δ e = ∂ ˉ λ + ∂ λ ˉ − L ϑ ( λ ) e . On the level of L ∞ L_\infty L ∞ algebras this is the isomorphism f : L → L ′ f:L\to L' f : L → L ′ , a notion to be reviewed below, with f 1 = 1 f_1=1 f 1 = 1 and f 2 ( λ , e ) = 1 2 λ ∙ e f_2(\lambda,e)=\frac{1}{2}\lambda\bullet e f 2 ( λ , e ) = 2 1 λ ∙ e . The transformed L ∞ L_\infty L ∞ brackets, denoted by a prime, include ℓ 2 ′ ( λ , e ) = L ϑ ( λ ) e \ell'_2(\lambda,e)=\mathcal{L}_{\vartheta(\lambda)}e ℓ 2 ′ ( λ , e ) = L ϑ ( λ ) e , ℓ 2 ′ ( λ , λ ˉ ) = L ϑ ( λ ) λ ˉ \ell'_2(\lambda,\bar{\lambda})=\mathcal{L}_{\vartheta(\lambda)}\bar{\lambda} ℓ 2 ′ ( λ , λ ˉ ) = L ϑ ( λ ) λ ˉ , while ℓ 2 ′ ( χ , e ) \ell'_2(\chi,e) ℓ 2 ′ ( χ , e ) and ℓ 3 ′ ( λ 1 , λ 2 , e ) \ell'_3(\lambda_1,\lambda_2,e) ℓ 3 ′ ( λ 1 , λ 2 , e ) vanish. The price is a ( 0 , 1 ) (0,1) ( 0 , 1 ) -component of the gauge bracket,
ℓ 2 ′ ( λ 1 , λ 2 ) = [ λ 1 , λ 2 ] C + 1 2 ( ι ϑ ( λ 2 ) ∂ ˉ λ 1 − ι ϑ ( λ 1 ) ∂ ˉ λ 2 ) − 1 2 ( ι ∂ ˉ ϑ ( λ 2 ) λ 1 − ι ∂ ˉ ϑ ( λ 1 ) λ 2 ) , (62) \ell'_2(\lambda_1,\lambda_2)=[\lambda_1,\lambda_2]_{\mathrm{C}}+\frac{1}{2}\left(\iota_{\vartheta(\lambda_2)}\bar{\partial}\lambda_1-\iota_{\vartheta(\lambda_1)}\bar{\partial}\lambda_2\right)-\frac{1}{2}\left(\iota_{\bar{\partial}\vartheta(\lambda_2)}\lambda_1-\iota_{\bar{\partial}\vartheta(\lambda_1)}\lambda_2\right), \tag*{(62)} ℓ 2 ′ ( λ 1 , λ 2 ) = [ λ 1 , λ 2 ] C + 2 1 ( ι ϑ ( λ 2 ) ∂ ˉ λ 1 − ι ϑ ( λ 1 ) ∂ ˉ λ 2 ) − 2 1 ( ι ∂ ˉ ϑ ( λ 2 ) λ 1 − ι ∂ ˉ ϑ ( λ 1 ) λ 2 ) , (62)
where the second and third terms are ( 0 , 1 ) (0,1) ( 0 , 1 ) -forms, with the contractions acting on the holomorphic indices.
Cyclic L ∞ L_\infty L ∞ algebra and BV master action
In order to derive the theory from an action we need, in addition, an inner product that is compatible with the brackets. The folded complex carries the inner product
⟨ ⟨ a , b ⟩ ⟩ = ⟨ ⟨ b , a ⟩ ⟩ : = ∫ X b ∧ a , (63) \langle\langle a,b\rangle\rangle=\langle\langle b,a\rangle\rangle:=\int_X b\wedge a, \tag*{(63)} ⟨⟨ a , b ⟩⟩ = ⟨⟨ b , a ⟩⟩ := ∫ X b ∧ a , (63)
for a a a one of χ , λ , λ ˉ , e \chi,\lambda,\bar{\lambda},e χ , λ , λ ˉ , e and b b b the corresponding one of χ ∗ , λ ∗ , λ ˉ ∗ , e ∗ \chi^*,\lambda^*,\bar{\lambda}^*,e^* χ ∗ , λ ∗ , λ ˉ ∗ , e ∗ , the antifield being placed first independently of the order of the arguments. It has degree − 3 -3 − 3 and is graded symmetric.6 An L ∞ L_\infty L ∞ algebra is cyclic with respect to ⟨ ⟨ ⋅ , ⋅ ⟩ ⟩ \langle\langle\cdot,\cdot\rangle\rangle ⟨⟨ ⋅ , ⋅ ⟩⟩ if
⟨ ⟨ x 0 , ℓ n ( x 1 , … , x n ) ⟩ ⟩ = ( − 1 ) n + n ( x 0 + x n ) + x n ( x 0 + ⋯ + x n − 1 ) ⟨ ⟨ x n , ℓ n ( x 0 , … , x n − 1 ) ⟩ ⟩ , (64) \langle\langle x_0,\ell_n(x_1,\ldots,x_n)\rangle\rangle=(-1)^{n+n(x_0+x_n)+x_n(x_0+\cdots+x_{n-1})}\langle\langle x_n,\ell_n(x_0,\ldots,x_{n-1})\rangle\rangle, \tag*{(64)} ⟨⟨ x 0 , ℓ n ( x 1 , … , x n )⟩⟩ = ( − 1 ) n + n ( x 0 + x n ) + x n ( x 0 + ⋯ + x n − 1 ) ⟨⟨ x n , ℓ n ( x 0 , … , x n − 1 )⟩⟩ , (64)
in the conventions of [43 ] . Together with the graded antisymmetry (51) of the ℓ n \ell_n ℓ n , this determines the behavior of ⟨ ⟨ x 0 , ℓ n ( x 1 , … , x n ) ⟩ ⟩ \langle\langle x_0,\ell_n(x_1,\ldots,x_n)\rangle\rangle ⟨⟨ x 0 , ℓ n ( x 1 , … , x n )⟩⟩ under all permutations of its n + 1 n+1 n + 1 arguments. The form is graded symmetric with respect to the shifted degrees ∣ x ∣ + 1 |x|+1 ∣ x ∣ + 1 , corresponding to the suspended so-called b b b -picture, which in turn explains the additional terms in the exponent of (64) . For n = 1 n=1 n = 1 , for instance, (64) states that ℓ 1 \ell_1 ℓ 1 is graded skew-adjoint, ⟨ ⟨ ℓ 1 x 0 , x 1 ⟩ ⟩ = \langle\langle\ell_1x_0,x_1\rangle\rangle= ⟨⟨ ℓ 1 x 0 , x 1 ⟩⟩ = − ( − 1 ) x 0 ⟨ ⟨ x 0 , ℓ 1 x 1 ⟩ ⟩ -(-1)^{x_0}\langle\langle x_0,\ell_1x_1\rangle\rangle − ( − 1 ) x 0 ⟨⟨ x 0 , ℓ 1 x 1 ⟩⟩ . For fields e e e in degree one we have ⟨ ⟨ ℓ 1 e , e ⟩ ⟩ = ⟨ ⟨ e , ℓ 1 e ⟩ ⟩ \langle\langle\ell_1e,e\rangle\rangle=\langle\langle e,\ell_1e\rangle\rangle ⟨⟨ ℓ 1 e , e ⟩⟩ = ⟨⟨ e , ℓ 1 e ⟩⟩ , as expected. More generally, cyclicity follows for ℓ 1 \ell_1 ℓ 1 defined in (3.21) by integration by parts, using (12) . The higher brackets of L L L , however, are not cyclic. For instance, cyclicity relates ⟨ ⟨ χ ∗ , ℓ 2 ( λ , χ ) ⟩ ⟩ \langle\langle\chi^*,\ell_2(\lambda,\chi)\rangle\rangle ⟨⟨ χ ∗ , ℓ 2 ( λ , χ )⟩⟩ to ⟨ ⟨ χ , ℓ 2 ( λ , χ ∗ ) ⟩ ⟩ \langle\langle\chi,\ell_2(\lambda,\chi^*)\rangle\rangle ⟨⟨ χ , ℓ 2 ( λ , χ ∗ )⟩⟩ , and after integration by parts the factor 1 2 \frac{1}{2} 2 1 in ℓ 2 ( λ , χ ) = 1 2 ι ϑ ( λ ) ∂ χ \ell_2(\lambda,\chi)=\frac{1}{2}\iota_{\vartheta(\lambda)}\partial\chi ℓ 2 ( λ , χ ) = 2 1 ι ϑ ( λ ) ∂ χ is not matched by ℓ 2 ( λ , χ ∗ ) = L ϑ ( λ ) χ ∗ \ell_2(\lambda,\chi^*)=\mathcal{L}_{\vartheta(\lambda)}\chi^* ℓ 2 ( λ , χ ∗ ) = L ϑ ( λ ) χ ∗ . The algebra L L L is, however, isomorphic to a cyclic one.
To show this we have to recall the notion of L ∞ L_\infty L ∞ morphisms. These are maps F : L → L ′ F:L\to L' F : L → L ′ consisting of graded antisymmetric multilinear maps F n : L ⊗ n → L ′ F_n:L^{\otimes n}\to L' F n : L ⊗ n → L ′ of degree 1 − n 1-n 1 − n , subject to relations that ensure compatibility with the brackets. In the lowest arities they state that F 1 F_1 F 1 is a chain map, ℓ 1 ′ F 1 = F 1 ℓ 1 \ell'_1F_1=F_1\ell_1 ℓ 1 ′ F 1 = F 1 ℓ 1 , and that F 1 F_1 F 1 intertwines the binary brackets up to a homotopy given by F 2 F_2 F 2 ,
ℓ 2 ′ ( F 1 x 1 , F 1 x 2 ) = F 1 ℓ 2 ( x 1 , x 2 ) + ℓ 1 ′ F 2 ( x 1 , x 2 ) + F 2 ( ℓ 1 x 1 , x 2 ) + ( − 1 ) x 1 F 2 ( x 1 , ℓ 1 x 2 ) . (65) \ell'_2(F_1x_1,F_1x_2)=F_1\ell_2(x_1,x_2)+\ell'_1F_2(x_1,x_2)+F_2(\ell_1x_1,x_2)+(-1)^{x_1}F_2(x_1,\ell_1x_2). \tag*{(65)} ℓ 2 ′ ( F 1 x 1 , F 1 x 2 ) = F 1 ℓ 2 ( x 1 , x 2 ) + ℓ 1 ′ F 2 ( x 1 , x 2 ) + F 2 ( ℓ 1 x 1 , x 2 ) + ( − 1 ) x 1 F 2 ( x 1 , ℓ 1 x 2 ) . (65)
The relation in arity three similarly expresses ℓ 3 ′ ( F 1 x 1 , F 1 x 2 , F 1 x 3 ) \ell'_3(F_1x_1,F_1x_2,F_1x_3) ℓ 3 ′ ( F 1 x 1 , F 1 x 2 , F 1 x 3 ) in terms of F 1 ℓ 3 F_1\ell_3 F 1 ℓ 3 , of F 3 F_3 F 3 composed with ℓ 1 \ell_1 ℓ 1 , and of compositions of F 2 F_2 F 2 with the binary brackets. A morphism with F n ≥ 2 = 0 F_{n\geq2}=0 F n ≥ 2 = 0 is called strict, and a morphism with invertible F 1 F_1 F 1 is an isomorphism. Given an L ∞ L_\infty L ∞ algebra L ′ L' L ′ and maps F n F_n F n with invertible F 1 F_1 F 1 , the morphism relations can be read as definitions of brackets ℓ n \ell_n ℓ n on the underlying complex of L L L , arity by arity, and these brackets then satisfy the L ∞ L_\infty L ∞ relations because those of L ′ L' L ′ do.
Proposition 5: Let F 1 = 1 F_1=1 F 1 = 1 , F n ≥ 3 = 0 F_{n\geq3}=0 F n ≥ 3 = 0 , and let the only non-vanishing components of F 2 F_2 F 2 be
F 2 ( λ , e ) = − 1 2 λ ∙ e , F_2(\lambda,e)=-\frac{1}{2}\lambda\bullet e, F 2 ( λ , e ) = − 2 1 λ ∙ e , F 2 ( λ , χ ∗ ) = ∂ λ ∧ c χ − 1 2 λ ∧ ∂ c χ , F 2 ( e , λ ˉ ∗ ) = 2 ∂ e ∧ c + e ∧ ∂ c , (66) F_2(\lambda,\chi^*)=\partial\lambda\wedge c_\chi-\frac{1}{2}\lambda\wedge\partial c_\chi,\qquad F_2(e,\bar{\lambda}^*)=2\partial e\wedge c+e\wedge\partial c, \tag*{(66)} F 2 ( λ , χ ∗ ) = ∂ λ ∧ c χ − 2 1 λ ∧ ∂ c χ , F 2 ( e , λ ˉ ∗ ) = 2 ∂ e ∧ c + e ∧ ∂ c , (66) where c = λ ˉ ∗ ⌟ ω − 1 ∈ Ω 0 , 2 c=\bar{\lambda}^*\lrcorner\omega^{-1}\in\Omega^{0,2} c = λ ˉ ∗ ┘ ω − 1 ∈ Ω 0 , 2 and c χ = χ ∗ ⌟ ω − 1 ∈ Ω 0 , 3 c_\chi=\chi^*\lrcorner\omega^{-1}\in\Omega^{0,3} c χ = χ ∗ ┘ ω − 1 ∈ Ω 0 , 3 . The L ∞ L_\infty L ∞ algebra L c y c L_{\mathrm{cyc}} L cyc on the folded complex for which F : L c y c → L F:L_{\mathrm{cyc}}\to L F : L cyc → L is an L ∞ L_\infty L ∞ isomorphism is cyclic with respect to (63) .
Explicitly, (65) gives the binary brackets of L c y c L_{\mathrm{cyc}} L cyc as
ℓ 2 c y c ( x 1 , x 2 ) = ℓ 2 ( x 1 , x 2 ) − ℓ 1 F 2 ( x 1 , x 2 ) − F 2 ( ℓ 1 x 1 , x 2 ) − ( − 1 ) x 1 F 2 ( x 1 , ℓ 1 x 2 ) , (67) \ell^{\mathrm{cyc}}_2(x_1,x_2)=\ell_2(x_1,x_2)-\ell_1F_2(x_1,x_2)-F_2(\ell_1x_1,x_2)-(-1)^{x_1}F_2(x_1,\ell_1x_2), \tag*{(67)} ℓ 2 cyc ( x 1 , x 2 ) = ℓ 2 ( x 1 , x 2 ) − ℓ 1 F 2 ( x 1 , x 2 ) − F 2 ( ℓ 1 x 1 , x 2 ) − ( − 1 ) x 1 F 2 ( x 1 , ℓ 1 x 2 ) , (67)
and the ternary bracket follows in the same way from the relation in arity three, while ℓ n ≥ 4 c y c = 0 \ell^{\mathrm{cyc}}_{n\geq4}=0 ℓ n ≥ 4 cyc = 0 . By construction, L c y c L_{\mathrm{cyc}} L cyc is an L ∞ L_\infty L ∞ algebra isomorphic to L L L : its brackets are defined by the morphism relations, as explained above, and they satisfy the L ∞ L_\infty L ∞ relations because those of L L L do. The non-trivial statement of Proposition 5 is that these brackets are cyclic. We have verified this by direct computation, i.e. by checking (64) for ℓ 2 c y c \ell^{\mathrm{cyc}}_2 ℓ 2 cyc and ℓ 3 c y c \ell^{\mathrm{cyc}}_3 ℓ 3 cyc on all combinations of fields. The map (66) itself can be found in the same way: making the most general ansatz for F 2 F_2 F 2 compatible with the degrees and with conservation of the antiholomorphic degree, the conditions (64) for the transported brackets are linear in the coefficients of the ansatz, and solving them gives (66) . No F 3 F_3 F 3 is possible by the same degree counting.
The component F 2 ( λ , e ) F_2(\lambda,e) F 2 ( λ , e ) is the field-dependent redefinition of λ ˉ \bar{\lambda} λ ˉ discussed at the end of section 3.2: composing F F F with the isomorphism f : L → L ′ f:L\to L' f : L → L ′ removes it, so that L c y c L_{\mathrm{cyc}} L cyc is related to L ′ L' L ′ by an isomorphism whose only components are the two λ ∗ \lambda^* λ ∗ -valued ones in (66) . Consequently, ℓ 2 c y c ( λ , e ) = L ϑ ( λ ) e \ell^{\mathrm{cyc}}_2(\lambda,e)=\mathcal{L}_{\vartheta(\lambda)}e ℓ 2 cyc ( λ , e ) = L ϑ ( λ ) e , as in the master action below, and ℓ 2 c y c ( λ 1 , λ 2 ) = ℓ 2 ( λ 1 , λ 2 ) \ell^{\mathrm{cyc}}_2(\lambda_1,\lambda_2)=\ell_2(\lambda_1,\lambda_2) ℓ 2 cyc ( λ 1 , λ 2 ) = ℓ 2 ( λ 1 , λ 2 ) , including the ( 0 , 1 ) (0,1) ( 0 , 1 ) -component in (62) . The other two components of F 2 F_2 F 2 take values in the λ ∗ \lambda^* λ ∗ -sector. They account for a factor 1 2 \frac{1}{2} 2 1 in the Lie derivative on χ ∗ \chi^* χ ∗ , which repairs the example above, for additional brackets ℓ 2 ( χ , χ ∗ ) \ell_2(\chi,\chi^*) ℓ 2 ( χ , χ ∗ ) , ℓ 2 ( λ ˉ , λ ˉ ∗ ) \ell_2(\bar{\lambda},\bar{\lambda}^*) ℓ 2 ( λ ˉ , λ ˉ ∗ ) and ℓ 2 ( λ , λ ˉ ∗ ) \ell_2(\lambda,\bar{\lambda}^*) ℓ 2 ( λ , λ ˉ ∗ ) with values in Ω 2 , 3 \Omega^{2,3} Ω 2 , 3 , for the different form of ℓ 2 ( λ , λ ∗ ) \ell_2(\lambda,\lambda^*) ℓ 2 ( λ , λ ∗ ) and ℓ 2 ( e , e ∗ ) \ell_2(e,e^*) ℓ 2 ( e , e ∗ ) , for the absence of ℓ 2 ( e , λ ˉ ∗ ) \ell_2(e,\bar{\lambda}^*) ℓ 2 ( e , λ ˉ ∗ ) , and for the cyclic partner ℓ 3 ( λ , λ , χ ∗ ) ∈ Ω 2 , 3 \ell_3(\lambda,\lambda,\chi^*)\in\Omega^{2,3} ℓ 3 ( λ , λ , χ ∗ ) ∈ Ω 2 , 3 of ℓ 3 ( λ , λ , λ ) \ell_3(\lambda,\lambda,\lambda) ℓ 3 ( λ , λ , λ ) .
Instead of listing the cyclic L ∞ L_\infty L ∞ brackets explicitly we encode them in the associated BV master action. Given a cyclic L ∞ L_\infty L ∞ algebra, the corresponding BV action is
S [ Φ ] = ∑ n ≥ 1 1 ( n + 1 ) ! ⟨ ⟨ Φ , ℓ n ( Φ , … , Φ ) ⟩ ⟩ (68) S[\Phi]=\sum_{n\geq1}\frac{1}{(n+1)!}\langle\!\langle\Phi,\ell_n(\Phi,\ldots,\Phi)\rangle\!\rangle \tag*{(68)} S [ Φ ] = n ≥ 1 ∑ ( n + 1 )! 1 ⟨ ⟨ Φ , ℓ n ( Φ , … , Φ )⟩ ⟩ (68)
where Φ \Phi Φ is the BV superfield of total degree one. Concretely,
Φ = χ + λ + λ ˉ + e + e ∗ + λ ∗ + λ ˉ ∗ + χ ∗ (69) \Phi=\chi+\lambda+\bar{\lambda}+e+e^*+\lambda^*+\bar{\lambda}^*+\chi^* \tag*{(69)} Φ = χ + λ + λ ˉ + e + e ∗ + λ ∗ + λ ˉ ∗ + χ ∗ (69)
where each component is a form of the bidegree given in § 3.2 , but with coefficient functions taking values in a Grassmann algebra, for instance λ = λ i ( z , z ˉ ) d z i \lambda=\lambda_i(z,\bar{z})\,dz^i λ = λ i ( z , z ˉ ) d z i with anticommuting λ i \lambda_i λ i . The coefficients carry the ghost number, defined as one minus the L ∞ L_\infty L ∞ degree: 2 for χ \chi χ , 1 for λ \lambda λ and λ ˉ \bar{\lambda} λ ˉ , 0 for e e e , and − 1 , − 2 , − 2 , − 3 -1,-2,-2,-3 − 1 , − 2 , − 2 , − 3 for e ∗ e^* e ∗ , λ ∗ \lambda^* λ ∗ , λ ˉ ∗ \bar{\lambda}^* λ ˉ ∗ , χ ∗ \chi^* χ ∗ , and their Grassmann parity is the ghost number mod 2. In finite dimensions this corresponds to writing Φ = ∑ a τ a ϕ a \Phi=\sum_a\tau_a\phi^a Φ = ∑ a τ a ϕ a with a basis τ a \tau_a τ a and graded coordinates ϕ a \phi^a ϕ a . Here the role of the label a a a is played by the form indices and the point of X X X , and the sum is the component expansion of the forms. The brackets and the inner product are extended to such Grassmann-valued forms by multilinearity, with a Koszul sign whenever a Grassmann-odd coefficient is moved past a form leg or past another odd coefficient. Since Φ \Phi Φ has total degree one, the graded antisymmetry (3.17) turns into ordinary symmetry when all arguments of ℓ n \ell_n ℓ n equal Φ \Phi Φ , so that ℓ n ( Φ , … , Φ ) \ell_n(\Phi,\ldots,\Phi) ℓ n ( Φ , … , Φ ) is in general non-zero. Expanding (3.33) in components then gives the action, the factor 1 / ( n + 1 ) ! 1/(n+1)! 1/ ( n + 1 )! cancelling against the multiplicities of the terms. Conversely, the action determines the brackets: by cyclicity ⟨ ⟨ x 0 , ℓ n ( x 1 , … , x n ) ⟩ ⟩ \langle\!\langle x_0,\ell_n(x_1,\ldots,x_n)\rangle\!\rangle ⟨ ⟨ x 0 , ℓ n ( x 1 , … , x n )⟩ ⟩ is obtained from the term of order n + 1 n+1 n + 1 in S S S by polarization. The L ∞ L_\infty L ∞ relations of a cyclic L ∞ L_\infty L ∞ algebra are equivalent to the classical master equation of S S S . By Propositions 4 and 5 , the action of L c y c L_{\mathrm{cyc}} L cyc therefore solves the classical master equation. In components it reads
S = ∫ X [ 1 2 e ∧ ∂ ˉ ∂ ˉ e − e ∗ ∧ ( ∂ ˉ λ ˉ + ∂ λ − L ϑ ( λ ) e ) + λ ∗ ∧ ( ∂ χ + 1 2 [ λ , λ ] C ) + λ ˉ ∗ ∧ ( ∂ ˉ χ + L ϑ ( λ ) λ ˉ + 1 2 ℓ 2 ′ ( λ , λ ) ∣ Ω 0 , 1 ) + χ ∗ ∧ ( 1 2 L ϑ ( λ ) ∂ χ + 1 6 T ( λ , λ , λ ) ) ] + 1 6 ∫ X ω ∧ ( ( A ∧ A ∧ A ∧ A ) ⌟ ω ) , (70) \begin{aligned}
S={}&\int_X\left[\frac{1}{2}e\wedge\bar{\partial}\bar{\partial}e-e^*\wedge\left(\bar{\partial}\bar{\lambda}+\partial\lambda-\mathcal{L}_{\vartheta(\lambda)}e\right)+\lambda^*\wedge\left(\partial\chi+\frac{1}{2}[\lambda,\lambda]_{\mathrm{C}}\right)\right.\\
&\left.+\bar{\lambda}^*\wedge\left(\bar{\partial}\chi+\mathcal{L}_{\vartheta(\lambda)}\bar{\lambda}+\frac{1}{2}\ell'_2(\lambda,\lambda)\big|_{\Omega^{0,1}}\right)+\chi^*\wedge\left(\frac{1}{2}\mathcal{L}_{\vartheta(\lambda)}\partial\chi+\frac{1}{6}T(\lambda,\lambda,\lambda)\right)\right]\\
&+\frac{1}{6}\int_X\omega\wedge\left((A\wedge A\wedge A\wedge A)\lrcorner\omega\right), \tag*{(70)}
\end{aligned} S = ∫ X [ 2 1 e ∧ ∂ ˉ ∂ ˉ e − e ∗ ∧ ( ∂ ˉ λ ˉ + ∂ λ − L ϑ ( λ ) e ) + λ ∗ ∧ ( ∂ χ + 2 1 [ λ , λ ] C ) + λ ˉ ∗ ∧ ( ∂ ˉ χ + L ϑ ( λ ) λ ˉ + 2 1 ℓ 2 ′ ( λ , λ ) Ω 0 , 1 ) + χ ∗ ∧ ( 2 1 L ϑ ( λ ) ∂ χ + 6 1 T ( λ , λ , λ ) ) ] + 6 1 ∫ X ω ∧ ( ( A ∧ A ∧ A ∧ A ) ┘ ω ) , (70)
where the Grassmann parameters are ordered as written, A = ϑ ( e ) A=\vartheta(e) A = ϑ ( e ) is the Beltrami differential of § 3.2 , and ℓ 2 ′ ( λ , λ ) ∣ Ω 0 , 1 \ell'_2(\lambda,\lambda)|_{\Omega^{0,1}} ℓ 2 ′ ( λ , λ ) ∣ Ω 0 , 1 is the ( 0 , 1 ) (0,1) ( 0 , 1 ) -form part of the bracket (3.27). With the conventions of § 2 one has, up to a total derivative, ∫ X ω ∧ ( ( A ∧ A ∧ A ∧ A ) ⌟ ω ) = − ∫ X e ∧ { ϑ ( e ) , ϑ ( e ) } ω \int_X\omega\wedge((A\wedge A\wedge A\wedge A)\lrcorner\omega)=-\int_Xe\wedge\{\vartheta(e),\vartheta(e)\}\omega ∫ X ω ∧ (( A ∧ A ∧ A ∧ A ) ┘ ω ) = − ∫ X e ∧ { ϑ ( e ) , ϑ ( e )} ω . The structure is the familiar one: the classical action 1 2 ∫ e ∧ ∂ ˉ ∂ e + 1 6 ∫ ω ∧ ( ( A ∧ A ∧ A ∧ A ) ⌟ ω ) \frac{1}{2}\int e\wedge\bar{\partial}\partial e+\frac{1}{6}\int\omega\wedge((A\wedge A\wedge A\wedge A)\lrcorner\omega) 2 1 ∫ e ∧ ∂ ˉ ∂ e + 6 1 ∫ ω ∧ (( A ∧ A ∧ A ∧ A ) ┘ ω ) , which is precisely the local action, followed by the antifields multiplying the gauge transformations of the corresponding fields, in the Lie derivative form δ e = ∂ ˉ λ ˉ + ∂ λ − L ϑ ( λ ) e \delta e=\bar{\partial}\bar{\lambda}+\partial\lambda-\mathcal{L}_{\vartheta(\lambda)}e δ e = ∂ ˉ λ ˉ + ∂ λ − L ϑ ( λ ) e discussed at the end of § 3.2 , and the terms encoding the reducibility.
Reduction to three dimensions
We now reduce the theory to the real slice of X = C 3 X = \mathbb{C}^{3} X = C 3 defined by z i = z ˉ ı ˉ z^{i} = \bar{z}^{\bar{\imath}} z i = z ˉ ˉ . The fields are thus independent of the imaginary parts y i y^{i} y i in (8) . Furthermore, we rescale the real parts of the coordinates by a factor of 2 so that x i = z i + z ˉ ı ˉ x^{i} = z^{i} + \bar{z}^{\bar{\imath}} x i = z i + z ˉ ˉ , and
∂ i = ∂ ˉ ı ˉ = ∂ ∂ x i . (71) \partial_{i} = \bar{\partial}_{\bar{\imath}} = \frac{\partial}{\partial x^{i}} . \tag*{(71)} ∂ i = ∂ ˉ ˉ = ∂ x i ∂ . (71)
Holomorphic and antiholomorphic indices then both become indices of R 3 \mathbb{R}^{3} R 3 , so that e i ȷ ˉ e_{i\bar{\jmath}} e i ˉ becomes a 3 × 3 3 \times3 3 × 3 matrix e i j e_{ij} e ij . Moreover, the 3-form ω i j k \omega_{ijk} ω ij k reduces to the Levi-Civita symbol ϵ i j k \epsilon_{ijk} ϵ ij k of R 3 \mathbb{R}^{3} R 3 . We normalize the volume factor of the imaginary parts y i y^{i} y i to one, i.e. ∫ X f d z 1 ∧ d z 2 ∧ d z 3 ∧ d z ˉ 1 ˉ ∧ d z ˉ 2 ˉ ∧ d z ˉ 3 ˉ = ∫ d 3 x f \int_{X} f\, dz^{1} \wedge dz^{2} \wedge dz^{3} \wedge d\bar{z}^{\bar{1}} \wedge d\bar{z}^{\bar{2}} \wedge d\bar{z}^{\bar{3}} = \int d^{3}x\, f ∫ X f d z 1 ∧ d z 2 ∧ d z 3 ∧ d z ˉ 1 ˉ ∧ d z ˉ 2 ˉ ∧ d z ˉ 3 ˉ = ∫ d 3 x f , and from now on write ∂ i \partial_{i} ∂ i for ∂ / ∂ x i \partial/\partial x^{i} ∂ / ∂ x i . The classical action 1 2 ∫ e ∧ ∂ ∂ ˉ e + 1 6 ∫ ω ∧ ( ( A ∧ A ∧ A ) ⌟ ω ) \frac{1}{2}\int e \wedge\partial\bar{\partial}e + \frac{1}{6}\int\omega\wedge((A \wedge A \wedge A)\mathbin{\lrcorner}\omega) 2 1 ∫ e ∧ ∂ ∂ ˉ e + 6 1 ∫ ω ∧ (( A ∧ A ∧ A ) ┘ ω ) then reduces to
S = − ∫ d 3 x ( 1 2 ϵ i k m ϵ j l n e i j ∂ k ∂ l e m n + det A ) , A j i = ϵ i k l ∂ k e l j , (72) S = - \int d^{3}x \left( \frac{1}{2}\epsilon^{ikm}\epsilon^{jln}e_{ij}\partial_{k}\partial_{l}e_{mn} + \det A \right), \qquad A^{i}_{\ j} = \epsilon^{ikl}\partial_{k}e_{lj} , \tag*{(72)} S = − ∫ d 3 x ( 2 1 ϵ ik m ϵ j l n e ij ∂ k ∂ l e mn + det A ) , A j i = ϵ ik l ∂ k e l j , (72)
where A A A is the Beltrami differential ϑ ( e ) \vartheta(e) ϑ ( e ) . The gauge transformations in (70) become
δ e i j = ∂ i λ ˉ j − ∂ j λ i − ( X k ∂ k e i j + ∂ i X k e k j ) , X i = ϑ ( λ ) i = ϵ i j k ∂ j λ k . (73) \delta e_{ij} = \partial_{i}\bar{\lambda}_{j} - \partial_{j}\lambda_{i} - \left(X^{k}\partial_{k}e_{ij} + \partial_{i}X^{k}e_{kj}\right), \qquad X^{i} = \vartheta(\lambda)^{i} = \epsilon^{ijk}\partial_{j}\lambda_{k} . \tag*{(73)} δ e ij = ∂ i λ ˉ j − ∂ j λ i − ( X k ∂ k e ij + ∂ i X k e k j ) , X i = ϑ ( λ ) i = ϵ ij k ∂ j λ k . (73)
The field-dependent term is the Lie derivative along the divergence-free vector field X X X , acting on the first index of e e e . Note that with ∂ k X k = 0 \partial_{k}X^{k} = 0 ∂ k X k = 0 one computes
δ det A = − ∂ k ( X k det A ) , (74) \delta\det A = -\partial_{k}\left(X^{k}\det A\right) , \tag*{(74)} δ det A = − ∂ k ( X k det A ) , (74)
which in turn implies that the cubic term is invariant under the field-dependent transformations, up to a total derivative, for arbitrary e e e . In fact, (72) is exactly gauge invariant under (73) , as expected from the gauge invariance of the full Kodaira–Spencer theory.
We next decompose e e e into its symmetric and antisymmetric parts,
e i j = h i j + b i j , h i j = h ( i j ) , b i j = : ϵ i j k v k . (75) e_{ij} = h_{ij} + b_{ij}, \qquad h_{ij} = h_{(ij)}, \qquad b_{ij} =: \epsilon_{ijk}v^{k} . \tag*{(75)} e ij = h ij + b ij , h ij = h ( ij ) , b ij =: ϵ ij k v k . (75)
The operator ϵ i k m ϵ j l n ∂ k ∂ l \epsilon^{ikm}\epsilon^{jln}\partial_{k}\partial_{l} ϵ ik m ϵ j l n ∂ k ∂ l is invariant under the simultaneous exchange i ↔ j i \leftrightarrow j i ↔ j , m ↔ n m \leftrightarrow n m ↔ n , so that the terms mixing h h h and b b b in the quadratic action are total derivatives. Using that the linearized Einstein tensor G i j G^{ij} G ij in three dimensions can be written as [44 ]
G i j ( h ) = 1 2 ϵ i k m ϵ j l n ∂ k ∂ l h m n , (76) G^{ij}(h) = \frac{1}{2}\epsilon^{ikm}\epsilon^{jln}\partial_{k}\partial_{l}h_{mn} , \tag*{(76)} G ij ( h ) = 2 1 ϵ ik m ϵ j l n ∂ k ∂ l h mn , (76)
and introducing the field strength H = d b H = db H = d b , whose only independent component is H 123 = ∂ k v k H_{123} = \partial_{k}v^{k} H 123 = ∂ k v k , we obtain
S 2 = − ∫ d 3 x h i j G i j ( h ) + ∫ d 3 x ( ∂ k v k ) 2 = − ∫ d 3 x h i j G i j ( h ) + 1 6 ∫ d 3 x H i j k H i j k . (77) \begin{aligned}
S_{2} &= - \int d^{3}x\, h^{ij}G_{ij}(h) + \int d^{3}x\,(\partial_{k}v^{k})^{2} \\
&= - \int d^{3}x\, h^{ij}G_{ij}(h) + \frac{1}{6}\int d^{3}x\,H_{ijk}H^{ijk} . \tag*{(77)}
\end{aligned} S 2 = − ∫ d 3 x h ij G ij ( h ) + ∫ d 3 x ( ∂ k v k ) 2 = − ∫ d 3 x h ij G ij ( h ) + 6 1 ∫ d 3 x H ij k H ij k . (77)
The first term is the Fierz–Pauli action, twice the quadratic part of ∫ g R \int\sqrt{g}R ∫ g R for g i j = δ i j + h i j g_{ij} = \delta_{ij} + h_{ij} g ij = δ ij + h ij . The second term is the kinetic term of a Kalb–Ramond field, which relative to the combination R − 1 12 H 2 R - \frac{1}{12}H^{2} R − 12 1 H 2 of non-topological string theory has the opposite sign, but we are in Euclidean signature, and a two-form is anyhow not propagating in three dimensions. The linearized gauge transformations read
δ h i j = 2 ∂ ( i ξ j ) , ξ = 1 2 ( λ ˉ − λ ) , \delta h_{ij}=2\partial_{(i}\xi_{j)},\qquad\xi=\frac{1}{2}(\bar{\lambda}-\lambda), δ h ij = 2 ∂ ( i ξ j ) , ξ = 2 1 ( λ ˉ − λ ) ,
δ b i j = 2 ∂ [ i Λ j ] , Λ = 1 2 ( λ ˉ + λ ) , (78) \delta b_{ij}=2\partial_{[i}\Lambda_{j]},\qquad\Lambda=\frac{1}{2}(\bar{\lambda}+\lambda), \tag*{(78)} δ b ij = 2 ∂ [ i Λ j ] , Λ = 2 1 ( λ ˉ + λ ) , (78)
which are the linearized diffeomorphisms and B B B -field gauge transformations, with λ \lambda λ and λ ˉ \bar{\lambda} λ ˉ entering precisely as in double field theory [13 ] . In contrast to standard double field theory, no dilaton is needed. On-shell, the b b b field just gives an integration constant: varying the action with respect to v k v^k v k yields ∂ k ( ∂ l v l ) = 0 \partial_k(\partial_l v^l)=0 ∂ k ( ∂ l v l ) = 0 , which implies v k = m x k v^k=mx^k v k = m x k , with m m m constant, up to a pure gauge part δ v k = ϵ k l m ∂ l Λ m \delta v^k=\epsilon^{klm}\partial_l\Lambda_m δ v k = ϵ k l m ∂ l Λ m that we set to zero.
Let us now turn to the cubic interactions. Remarkably, they are not those of Einstein–Hilbert gravity, because the vertex has three derivatives and breaks parity. Moreover, the gauge transformations act through the divergence-free vector fields X X X rather than through the parameter ξ \xi ξ of (78) . Such deformations are not covered by uniqueness results as given in [45 ] , which assume parity invariance and at most two derivatives. However, a deformation of the type found here was constructed by Boulanger and Gualtieri [31 ] . To compare with [31 ] , we set b = 0 b=0 b = 0 . Then − det A = 4 a 0 -\det A=4a_0 − det A = 4 a 0 , with
a 0 = 1 3 ϵ i j k δ l m ∂ [ l h n ] i ∂ [ n h p ] j ∂ [ m h p ] k , (79) a_0=\frac{1}{3}\epsilon^{ijk}\delta^{lm}\partial_{[l}h_{n]i}\partial_{[n}h_{p]j}\partial_{[m}h_{p]k}, \tag*{(79)} a 0 = 3 1 ϵ ij k δ l m ∂ [ l h n ] i ∂ [ n h p ] j ∂ [ m h p ] k , (79)
which is precisely the vertex given in eq. (1.3) of [31 ] .
To analyze the consistency of the theory beyond cubic order, let us inspect the full gauge transformations for non-vanishing h h h and b b b . Inserting λ = Λ − ξ \lambda=\Lambda-\xi λ = Λ − ξ and λ ˉ = Λ + ξ \bar{\lambda}=\Lambda+\xi λ ˉ = Λ + ξ into (73) and decomposing into symmetric and antisymmetric parts, we find
δ h i j = 2 ∂ ( i ξ j ) − X k ∂ k h i j − ∂ ( i X k h j ) k − ∂ ( i X k b ∣ k ∣ j ) , \delta h_{ij}=2\partial_{(i}\xi_{j)}-X^k\partial_kh_{ij}-\partial_{(i}X^k h_{j)k}-\partial_{(i}X^k b_{|k|j)}, δ h ij = 2 ∂ ( i ξ j ) − X k ∂ k h ij − ∂ ( i X k h j ) k − ∂ ( i X k b ∣ k ∣ j ) ,
δ b i j = 2 ∂ [ i Λ j ] − X k ∂ k b i j − ∂ [ i X k h j ] k − ∂ [ i X k b ∣ k ∣ j ] , (80) \delta b_{ij}=2\partial_{[i}\Lambda_{j]}-X^k\partial_kb_{ij}-\partial_{[i}X^k h_{j]k}-\partial_{[i}X^k b_{|k|j]}, \tag*{(80)} δ b ij = 2 ∂ [ i Λ j ] − X k ∂ k b ij − ∂ [ i X k h j ] k − ∂ [ i X k b ∣ k ∣ j ] , (80)
with X i = ϵ i j k ∂ j ( Λ k − ξ k ) X^i=\epsilon^{ijk}\partial_j(\Lambda_k-\xi_k) X i = ϵ ij k ∂ j ( Λ k − ξ k ) and (anti-)symmetrization with unit weight. Setting b = 0 b=0 b = 0 is not consistent with these gauge transformations, because the transformation of b b b contains the term − ∂ [ i X k h j ] k -\partial_{[i}X^k h_{j]k} − ∂ [ i X k h j ] k , and so a b b b is necessarily generated, even if initially it is set to zero. (Moreover, this term cannot be removed by a parameter redefinition: for general h h h it is not of the form ∂ [ i Λ j ] ′ \partial_{[i}\Lambda'_{j]} ∂ [ i Λ j ] ′ .)
Up to cubic order, however, it is easy to see that the truncation is consistent. For b = 0 b=0 b = 0 and Λ = 0 \Lambda=0 Λ = 0 , (80) reduces to
δ h i j = 2 ∂ ( i ξ j ) − X k ∂ k h i j − ∂ ( i X k h j ) k , X i = − ϵ i j k ∂ j ξ k , (81) \delta h_{ij}=2\partial_{(i}\xi_{j)}-X^k\partial_kh_{ij}-\partial_{(i}X^k h_{j)k},\qquad X^i=-\epsilon^{ijk}\partial_j\xi_k, \tag*{(81)} δ h ij = 2 ∂ ( i ξ j ) − X k ∂ k h ij − ∂ ( i X k h j ) k , X i = − ϵ ij k ∂ j ξ k , (81)
which agrees with eq. (4.37) in [31 ] , up to a field-dependent redefinition of ξ \xi ξ and the normalization of the coupling. Beyond cubic order, the truncated theory is not gauge invariant: the variation of its cubic vertex under the field-dependent part of (81) does not vanish, not even up to equations of motion and total derivatives.
Let us finally sketch how this infinite series is related to integrating out b b b . In terms of h h h and b b b or v v v , cf. (75) , the Beltrami differential reads
A j i = M j i + δ j i ∂ k v k − ∂ j v i , M j i = ϵ i k l ∂ k h l j . (82) A^{i}_{j}=M^{i}_{j}+\delta^{i}_{j}\partial_kv^k-\partial_jv^i,\qquad M^{i}_{j}=\epsilon^{ikl}\partial_kh_{lj}. \tag*{(82)} A j i = M j i + δ j i ∂ k v k − ∂ j v i , M j i = ϵ ik l ∂ k h l j . (82)
Expanding the vertex in powers of v v v , we obtain
det A = det M + C i j ( δ j i ∂ k v k − ∂ j v i ) + O ( v 2 ) , C i j = ∂ det M ∂ M j i = 1 2 ϵ i k l ϵ j m n M m k M n l , (83) \det A = \det M + C^{j}_{i}(\delta^{i}_{j}\partial_{k}v^{k}-\partial_{j}v^{i})+\mathcal{O}(v^{2}), \qquad C^{j}_{i}=\frac{\partial\det M}{\partial M^{i}_{j}}=\frac{1}{2}\epsilon_{ikl}\epsilon^{jmn}M^{k}_{m}M^{l}_{n}, \tag*{(83)} det A = det M + C i j ( δ j i ∂ k v k − ∂ j v i ) + O ( v 2 ) , C i j = ∂ M j i ∂ det M = 2 1 ϵ ik l ϵ j mn M m k M n l , (83)
where we defined the cofactor matrix C C C of M M M , which obeys C i j M k i = det M δ k j C^{j}_{i}M^{i}_{k}=\det M\,\delta^{j}_{k} C i j M k i = det M δ k j and is quadratic in h h h . Up to total derivatives, the cubic term linear in v v v is
S 3 ∣ O ( v ) = − ∫ d 3 x v i J i , J i = ∂ j C i j − ∂ i C j j . (84) S_{3}|_{\mathcal{O}(v)}=-\int d^{3}x\,v^{i}J_{i}, \qquad J_{i}=\partial_{j}C^{j}_{i}-\partial_{i}C^{j}_{j}. \tag*{(84)} S 3 ∣ O ( v ) = − ∫ d 3 x v i J i , J i = ∂ j C i j − ∂ i C j j . (84)
In the free theory, ∂ k v k = 3 m \partial_{k}v^{k}=3m ∂ k v k = 3 m is constant and the transverse part of v v v is pure gauge, so that integrating out v v v leaves the Fierz–Pauli action plus a constant. At first order in the coupling, v v v takes its free value, and for m = 0 m=0 m = 0 we obtain the Boulanger–Gualtieri vertex. (A non-zero m m m contributes − 2 m ∫ C i i -2m\int C^{i}_{i} − 2 m ∫ C i i , which one may verify to be proportional to the Fierz–Pauli action and hence merely rescales it.) At second order, we obtain the field equation 2 ∂ i ( ∂ k v k ) = − J i 2\partial_{i}(\partial_{k}v^{k})=-J_{i} 2 ∂ i ( ∂ k v k ) = − J i , with J J J defined in (3.49). We solve for v v v , upon fixing the gauge for its transverse part, which yields ∂ k v k = − 1 2 □ − 1 ∂ i J i \partial_{k}v^{k}=-\frac{1}{2}\square^{-1}\partial^{i}J_{i} ∂ k v k = − 2 1 □ − 1 ∂ i J i for m = 0 m=0 m = 0 . Next, we eliminate v i v^{i} v i in the action terms (3.42) and (3.49), where for the latter we can contract 2 ∂ i ( ∂ k v k ) = − J i 2\partial_{i}(\partial_{k}v^{k})=-J_{i} 2 ∂ i ( ∂ k v k ) = − J i with v i v^{i} v i , to obtain the quartic term
S 4 = − 1 4 ∫ d 3 x ( □ − 1 ∂ i J i ) 2 . (85) S_{4}=-\frac{1}{4}\int d^{3}x\,(\square^{-1}\partial^{i}J_{i})^{2}. \tag*{(85)} S 4 = − 4 1 ∫ d 3 x ( □ − 1 ∂ i J i ) 2 . (85)
The above action is non-local, but its non-localities can be removed by field redefinitions as follows. We first note with (76) that the divergence of the cofactor matrix is proportional to the free field equation:
∂ j C i j = ϵ i k l ϵ j m n ∂ j M m k M n l = 2 ϵ i k l G k n ( h ) M n l . (86) \partial_{j}C^{j}_{i}=\epsilon_{ikl}\epsilon^{jmn}\partial_{j}M^{k}_{m}M^{l}_{n}=2\epsilon_{ikl}G^{kn}(h)M^{l}_{n}. \tag*{(86)} ∂ j C i j = ϵ ik l ϵ j mn ∂ j M m k M n l = 2 ϵ ik l G k n ( h ) M n l . (86)
The current (3.49) thus takes the form J i = − ∂ i C j j + 2 ϵ i k l G k n ( h ) M n l J_{i}=-\partial_{i}C^{j}_{j}+2\epsilon_{ikl}G^{kn}(h)M^{l}_{n} J i = − ∂ i C j j + 2 ϵ ik l G k n ( h ) M n l ,7 and the quartic term (3.50) becomes
S 4 = − 1 4 ∫ d 3 x ( C j j ) 2 + ∫ d 3 x C j j N − ∫ d 3 x N 2 , N = □ − 1 ∂ i ( ϵ i k l G k n ( h ) M n l ) . (87) S_{4}=-\frac{1}{4}\int d^{3}x\,(C^{j}_{j})^{2}+\int d^{3}x\,C^{j}_{j}N-\int d^{3}x\,N^{2}, \qquad N=\square^{-1}\partial^{i}(\epsilon_{ikl}G^{kn}(h)M^{l}_{n}). \tag*{(87)} S 4 = − 4 1 ∫ d 3 x ( C j j ) 2 + ∫ d 3 x C j j N − ∫ d 3 x N 2 , N = □ − 1 ∂ i ( ϵ ik l G k n ( h ) M n l ) . (87)
The first term is local. The second and third term are non-local but can be written as ∫ d 3 x G k n ( h ) Y k n \int d^{3}x\,G^{kn}(h)Y_{kn} ∫ d 3 x G k n ( h ) Y k n , proportional to the free field equation, and hence be removed by a field redefinition h k n → h k n + 1 2 Y k n h_{kn}\rightarrow h_{kn}+\frac{1}{2}Y_{kn} h k n → h k n + 2 1 Y k n . These field redefinitions are non-local and generate further terms at higher order. Up to such redefinitions, integrating out b b b therefore yields, to second order in the coupling, the truncated theory S ∣ b = 0 S|_{b=0} S ∣ b = 0 together with the local quartic term − 1 4 ∫ d 3 x ( C j j ) 2 -\frac{1}{4}\int d^{3}x\,(C^{j}_{j})^{2} − 4 1 ∫ d 3 x ( C j j ) 2 . Integrating out b b b at higher orders will produce an infinite series, and it would be interesting to explore whether this reproduces the second-order formulation of [31 ] to all orders.
Double copy interpretation
In this section we show that the L ∞ L_{\infty} L ∞ algebra of section 3.2 is built, up to an explicit isomorphism, from two ‘single copies’: the Dolbeault algebra of antiholomorphic forms on the one hand, and the Roytenberg–Weinstein L ∞ L_{\infty} L ∞ algebra on holomorphic forms on the other, with the latter being mapped to the Lie algebra of holomorphic divergence-free vector fields by the embedding tensor. All L ∞ L_{\infty} L ∞ relations then follow from the relations of the single copies.
The single copies
The first single copy is the Dolbeault algebra
K = ( Ω 0 , ∙ , ∂ ˉ , ∧ ) , (88) \mathcal{K} = \left(\Omega^{0,\bullet}, \bar{\partial}, \wedge\right), \tag*{(88)} K = ( Ω 0 , ∙ , ∂ ˉ , ∧ ) , (88)
with Ω 0 , q \Omega^{0,q} Ω 0 , q in degree q q q , which is a differential graded commutative algebra. The subspace
K + : = Ω 0 , 2 ⊕ Ω 0 , 3 (89) \mathcal{K}^{+} := \Omega^{0,2} \oplus\Omega^{0,3} \tag*{(89)} K + := Ω 0 , 2 ⊕ Ω 0 , 3 (89)
is a differential ideal: it is mapped to itself by ∂ ˉ \bar{\partial} ∂ ˉ and by multiplication with arbitrary elements of K \mathcal{K} K . We write K − = Ω 0 , 0 ⊕ Ω 0 , 1 \mathcal{K}^{-} = \Omega^{0,0} \oplus\Omega^{0,1} K − = Ω 0 , 0 ⊕ Ω 0 , 1 , so that K = K − ⊕ K + \mathcal{K} = \mathcal{K}^{-} \oplus\mathcal{K}^{+} K = K − ⊕ K + as a vector space, and we have the isomorphism K / K + ≅ K − \mathcal{K}/\mathcal{K}^{+} \cong\mathcal{K}^{-} K / K + ≅ K − .
The second single copy is the Roytenberg–Weinstein algebra L R W L_{\mathrm{RW}} L RW of (45) , with brackets (46) . For the following construction it is important to display the dg Lie algebra into which L R W L_{\mathrm{RW}} L RW is mapped via the embedding tensor. It is most familiar as the dg Lie algebra of holomorphic vector fields and functions, Γ ( T 1 , 0 ) → div ω O \Gamma(T^{1,0}) \xrightarrow{\operatorname{div}_{\omega}} \mathcal{O} Γ ( T 1 , 0 ) div ω O , with the vector fields in degree 0, the functions in degree 1, the divergence div ω X = ∂ i X i \operatorname{div}_{\omega} X = \partial_i X^i div ω X = ∂ i X i as differential, and the brackets
[ X , Y ] = ( X j ∂ j Y i − Y j ∂ j X i ) ∂ i , [ X , f ] = X i ∂ i f , [ f , g ] = 0. (90) [X,Y] = \left(X^j\partial_jY^i - Y^j\partial_jX^i\right)\partial_i,\qquad[X,f] = X^i\partial_i f,\qquad[f,g] = 0. \tag*{(90)} [ X , Y ] = ( X j ∂ j Y i − Y j ∂ j X i ) ∂ i , [ X , f ] = X i ∂ i f , [ f , g ] = 0. (90)
For our purposes it is convenient to transport this algebra to forms with the help of ω \omega ω , using the isomorphisms X ↦ ι X ω ∈ Ω 2 , 0 X \mapsto\iota_X\omega\in\Omega^{2,0} X ↦ ι X ω ∈ Ω 2 , 0 and f ↦ f ω ∈ Ω 3 , 0 f \mapsto f\omega\in\Omega^{3,0} f ↦ f ω ∈ Ω 3 , 0 . Since ∂ ( ι X ω ) = ( div ω X ) ω \partial(\iota_X\omega) = (\operatorname{div}_{\omega} X)\omega ∂ ( ι X ω ) = ( div ω X ) ω by (20) , the divergence becomes the holomorphic differential, and we define
H = ( Ω 2 , 0 → ∂ Ω 3 , 0 ) , (91) \mathcal{H} = \left(\Omega^{2,0} \xrightarrow{\partial} \Omega^{3,0}\right), \tag*{(91)} H = ( Ω 2 , 0 ∂ Ω 3 , 0 ) , (91)
with holomorphic ( 2 , 0 ) (2,0) ( 2 , 0 ) -forms in degree 0 and holomorphic ( 3 , 0 ) (3,0) ( 3 , 0 ) -forms in degree 1. Writing β = ι X β ω \beta= \iota_{X_\beta}\omega β = ι X β ω for β ∈ Ω 2 , 0 \beta\in\Omega^{2,0} β ∈ Ω 2 , 0 and γ = f γ ω \gamma= f_\gamma\omega γ = f γ ω for γ ∈ Ω 3 , 0 \gamma\in\Omega^{3,0} γ ∈ Ω 3 , 0 , the brackets are
[ β 1 , β 2 ] = ι [ X β 1 , X β 2 ] ω , [ β , γ ] = X β ( f γ ) ω , [ γ 1 , γ 2 ] = 0. (92) [\beta_1,\beta_2] = \iota_{[X_{\beta_1},X_{\beta_2}]}\omega,\qquad[\beta,\gamma] = X_\beta(f_\gamma)\omega,\qquad[\gamma_1,\gamma_2] = 0. \tag*{(92)} [ β 1 , β 2 ] = ι [ X β 1 , X β 2 ] ω , [ β , γ ] = X β ( f γ ) ω , [ γ 1 , γ 2 ] = 0. (92)
Up to a sign, the bracket on Ω 2 , 0 \Omega^{2,0} Ω 2 , 0 is the Schouten bracket (21) transported by $\omega.
Proposition 6. H \mathcal{H} H is a dg Lie algebra.
Proof. Since (91) with (92) is isomorphic to the complex of vector fields and functions with (90) , it suffices to verify the claim for the latter. The Jacobi identity for three vector fields is that of the Lie bracket, and for two vector fields and a function it states that functions form a representation. That the divergence is a derivation, div ω [ X , Y ] = X ( div ω Y ) − Y ( div ω X ) \operatorname{div}_{\omega}[X,Y] = X(\operatorname{div}_{\omega}Y) - Y(\operatorname{div}_{\omega}X) div ω [ X , Y ] = X ( div ω Y ) − Y ( div ω X ) , follows since the terms ∂ i X j ∂ j Y i − ∂ i Y j ∂ j X i \partial_iX^j\partial_jY^i - \partial_iY^j\partial_jX^i ∂ i X j ∂ j Y i − ∂ i Y j ∂ j X i cancel. The derivation property on a vector field and a function holds trivially, since there is nothing in degree two. □
Since ι ϑ ( α ) ω = ∂ α \iota_{\vartheta(\alpha)}\omega= \partial\alpha ι ϑ ( α ) ω = ∂ α for α ∈ Ω 1 , 0 \alpha\in\Omega^{1,0} α ∈ Ω 1 , 0 by (36) , the embedding tensor ϑ \vartheta ϑ becomes, in this language, simply the holomorphic differential ∂ : Ω 1 , 0 → Ω 2 , 0 \partial: \Omega^{1,0} \to\Omega^{2,0} ∂ : Ω 1 , 0 → Ω 2 , 0 . The relation between the L ∞ L_\infty L ∞ algebra L R W L_{\mathrm{RW}} L RW and the dg Lie algebra H \mathcal{H} H is therefore encoded in the following diagram, which is the holomorphic de Rham complex split into two halves and then folded:
0 → Ω 2 , 0 → ∂ Ω 3 , 0 → h 0 ↑ ∂ ↑ 0 ↑ Ω 0 , 0 → ∂ Ω 1 , 0 → 0 → L R W \begin{CD}
0 @>>> \Omega^{2,0} @>{\partial}>> \Omega^{3,0} @>>> \mathfrak{h} \\
@A{0}AA @A{\partial}AA @A{0}AA \\
\Omega^{0,0} @>{\partial}>> \Omega^{1,0} @>>> 0 @>>> L_{\mathrm{RW}}
\end{CD} 0 0 ⏐ ↑ Ω 0 , 0 ∂ Ω 2 , 0 ∂ ⏐ ↑ Ω 1 , 0 ∂ Ω 3 , 0 0 ⏐ ↑ 0 h L RW
Proposition 7. The map Π : L R W → h \Pi: L_{\mathrm{RW}} \to\mathfrak{h} Π : L RW → h given by Π ( f ) = 0 \Pi(f) = 0 Π ( f ) = 0 and Π ( α ) = ∂ α \Pi(\alpha) = \partial\alpha Π ( α ) = ∂ α is a strict L ∞ L_\infty L ∞ morphism.
Proof: Π \Pi Π is a chain map by ∂ 2 = 0 \partial^2 = 0 ∂ 2 = 0 . Furthermore, Π \Pi Π intertwines the binary brackets: since X ∂ α = ϑ ( α ) X_{\partial\alpha} = \vartheta(\alpha) X ∂ α = ϑ ( α ) , the relation ∂ [ α 1 , α 2 ] C = [ ∂ α 1 , ∂ α 2 ] \partial[\alpha_1,\alpha_2]_{\mathrm{C}} = [\partial\alpha_1,\partial\alpha_2] ∂ [ α 1 , α 2 ] C = [ ∂ α 1 , ∂ α 2 ] is equivalent to ϑ ( [ α 1 , α 2 ] C ) = [ ϑ ( α 1 ) , ϑ ( α 2 ) ] \vartheta([\alpha_1,\alpha_2]_{\mathrm{C}}) = [\vartheta(\alpha_1),\vartheta(\alpha_2)] ϑ ([ α 1 , α 2 ] C ) = [ ϑ ( α 1 ) , ϑ ( α 2 )] , which holds by Proposition 1, and Π ℓ 2 ( α , f ) = 0 = [ ∂ α , Π f ] \Pi\ell_2(\alpha,f) = 0 = [\partial\alpha,\Pi f] Π ℓ 2 ( α , f ) = 0 = [ ∂ α , Π f ] . Finally, Π ℓ 3 = 0 \Pi\ell_3 = 0 Π ℓ 3 = 0 since ℓ 3 \ell_3 ℓ 3 takes values in Ω 0 , 0 = ker Π \Omega^{0,0} = \ker\Pi Ω 0 , 0 = ker Π , while h \mathfrak{h} h has no ternary bracket. □
Note that the non-strictness of L R W L_{\mathrm{RW}} L RW is invisible to Π \Pi Π , precisely because the image of its ternary bracket lies in the kernel of Π \Pi Π .
The double copy
Let us first recall how tensor products produce L ∞ L_\infty L ∞ algebras. Let A \mathcal{A} A be a graded commutative associative differential graded algebra, with differential d d d and product written as juxtaposition, a 1 a 2 ⋯ a n a_1a_2\cdots a_n a 1 a 2 ⋯ a n , and let L \mathcal{L} L be an L ∞ L_\infty L ∞ algebra with brackets ℓ n \ell_n ℓ n . Then A ⊗ L \mathcal{A}\otimes\mathcal{L} A ⊗ L is an L ∞ L_\infty L ∞ algebra with brackets
ℓ ^ 1 ( a ⊗ x ) : = d a ⊗ x + ( − 1 ) a a ⊗ ℓ 1 ( x ) , \widehat{\ell}_1(a\otimes x) := da\otimes x + (-1)^a a\otimes\ell_1(x), ℓ 1 ( a ⊗ x ) := d a ⊗ x + ( − 1 ) a a ⊗ ℓ 1 ( x ) ,
ℓ ^ n ( a 1 ⊗ x 1 , … , a n ⊗ x n ) : = ± a 1 ⋯ a n ⊗ ℓ n ( x 1 , … , x n ) , n ≥ 2 , (93) \widehat{\ell}_n(a_1\otimes x_1,\ldots,a_n\otimes x_n) := \pm a_1\cdots a_n\otimes\ell_n(x_1,\ldots,x_n), \qquad n\ge2, \tag*{(93)} ℓ n ( a 1 ⊗ x 1 , … , a n ⊗ x n ) := ± a 1 ⋯ a n ⊗ ℓ n ( x 1 , … , x n ) , n ≥ 2 , (93)
where the sign is the Koszul sign of moving ℓ n \ell_n ℓ n and the x i x_i x i past the a j a_j a j . Its L ∞ L_\infty L ∞ relations follow from those of L \mathcal{L} L together with d 2 = 0 d^2 = 0 d 2 = 0 , the Leibniz rule, associativity and commutativity in A \mathcal{A} A . Moreover, a strict morphism L → L ′ \mathcal{L}\to\mathcal{L}' L → L ′ induces a strict morphism A ⊗ L → A ⊗ L ′ \mathcal{A}\otimes\mathcal{L}\to\mathcal{A}\otimes\mathcal{L}' A ⊗ L → A ⊗ L ′ , and a differential ideal J J J of A \mathcal{A} A gives an L ∞ L_\infty L ∞ ideal J ⊗ L J\otimes\mathcal{L} J ⊗ L . This follows because every bracket with an argument in J ⊗ L J\otimes\mathcal{L} J ⊗ L has a factor in J J J .
Next, we will use the above statements for tensor products over C \mathbb{C} C , in the following sense. The holomorphic single copies L R W L_{\mathrm{RW}} L RW and h \mathfrak{h} h are taken to consist of objects whose coefficients depend holomorphically on z z z , and K \mathcal{K} K of antiholomorphic forms whose coefficients depend on z ˉ \bar{z} z ˉ . A priori, their tensor product over C \mathbb{C} C then consists of functions of independent variables z z z and z ˉ \bar{z} z ˉ , i.e. of functions on the complexification of X X X , which has twice the complex dimension of X X X . This is analogous to the standard double copy, where the fields depend on two sets of coordinates x x x and x ˉ \bar{x} x ˉ , one for each single copy. There, however, the brackets of the doubled theory satisfy the L ∞ L_\infty L ∞ relations only up to terms proportional to □ − □ ˉ \Box-\bar{\Box} □ − □ ˉ , so that the L ∞ L_\infty L ∞ algebra does not even exist on the doubled space. One has to impose the level-matching condition □ = □ ˉ \Box=\bar{\Box} □ = □ ˉ of string theory, which is solved by going to the undoubled space x = x ˉ x=\bar{x} x = x ˉ [33 ] . In the present case no such condition is needed, and the L ∞ L_\infty L ∞ algebra is already defined on the complexification, with z z z and z ˉ \bar{z} z ˉ independent. The real slice is needed, however, for the inner product, as the cycle over which the top form is integrated. A function that is holomorphic in both sets of coordinates is determined by its values on the real slice where z ˉ \bar{z} z ˉ is the complex conjugate of z z z and the derivatives ∂ i \partial_i ∂ i and ∂ ˉ ı ˉ \bar{\partial}_{\bar{\imath}} ∂ ˉ ˉ become the usual derivatives (9) . Restricting to the real slice is therefore compatible with all operations, and the tensor product describes real-analytic fields on X X X . General fields are obtained by completion, for which all relations continue to hold.
Let us then apply the above tensor product constructions to the single copies of the previous subsection. This gives the L ∞ L_\infty L ∞ algebra A = K ⊗ L R W A = \mathcal{K}\otimes L_{\mathrm{RW}} A = K ⊗ L RW , with brackets ℓ n A \ell_n^A ℓ n A as in (93) , and the dg Lie algebra B = K ⊗ h B = \mathcal{K}\otimes\mathfrak{h} B = K ⊗ h , with differential d B d_B d B and brackets [ ⋅ , ⋅ ] B [\cdot,\cdot]_B [ ⋅ , ⋅ ] B . Furthermore, Φ = 1 ⊗ Π : A → B \Phi= 1\otimes\Pi: A\to B Φ = 1 ⊗ Π : A → B defines a strict L ∞ L_\infty L ∞ morphism, and we have the ideals A + = K + ⊗ L R W A^+ = \mathcal{K}^+\otimes L_{\mathrm{RW}} A + = K + ⊗ L RW and B + = K + ⊗ h B^+ = \mathcal{K}^+\otimes\mathfrak{h} B + = K + ⊗ h . Concretely, A A A consists of the forms Ω p , q \Omega^{p,q} Ω p , q with p ≤ 1 p\leq1 p ≤ 1 , in degree p + q − 1 p+q-1 p + q − 1 , while B B B consists of the forms Ω p , q \Omega^{p,q} Ω p , q with p ≥ 2 p\geq2 p ≥ 2 , in degree p + q − 2 p+q-2 p + q − 2 .8 Every element of A A A is a sum of components of definite antiholomorphic degree, and we split it uniquely as a = a − + a + a=a^-+a^+ a = a − + a + , where a − a^- a − collects the components of antiholomorphic degree at most one and a + a^+ a + those of degree at least two. Thus a − ∈ A − : = K − ⊗ L R W a^-\in A^-:=\mathcal{K}^-\otimes L_{\mathrm{RW}} a − ∈ A − := K − ⊗ L RW and a + ∈ A + a^+\in A^+ a + ∈ A + , so that A = A − ⊕ A + A=A^-\oplus A^+ A = A − ⊕ A + as a vector space, and ( ⋅ ) ± (\cdot)^\pm ( ⋅ ) ± denote the projections onto the two summands. We use the same notation for B = B − ⊕ B + B=B^-\oplus B^+ B = B − ⊕ B + , with B − : = K − ⊗ h B^-:=\mathcal{K}^-\otimes\mathfrak{h} B − := K − ⊗ h . Note that A + A^+ A + and B + B^+ B + are ideals, while A − A^- A − and B − B^- B − are merely subspaces (a bracket of elements of A − A^- A − can have components in A + A^+ A + ). Since Φ \Phi Φ does not act on the antiholomorphic factor, it maps A ± A^\pm A ± into B ± B^\pm B ± .
The subspace A − A^- A − consists of χ \chi χ , λ \lambda λ , λ ˉ \bar{\lambda} λ ˉ and e e e , i.e. of the left half of the folded complex, while the ideal B + B^+ B + consists of the forms e ∗ e^* e ∗ , λ ∗ \lambda^* λ ∗ , λ ˉ ∗ \bar{\lambda}^* λ ˉ ∗ and χ ∗ \chi^* χ ∗ in Ω 2 , 2 \Omega^{2,2} Ω 2 , 2 , Ω 2 , 3 \Omega^{2,3} Ω 2 , 3 , Ω 3 , 2 \Omega^{3,2} Ω 3 , 2 and Ω 3 , 3 \Omega^{3,3} Ω 3 , 3 , i.e. of the right half of the folded complex. The double copy combines the two halves,
F = A − ⊕ B + = ( K − ⊗ L R W ) ⊕ ( K + ⊗ h ) , (94) \mathcal{F}=A^-\oplus B^+=(\mathcal{K}^-\otimes L_{\mathrm{RW}})\oplus(\mathcal{K}^+\otimes\mathfrak{h}), \tag*{(94)} F = A − ⊕ B + = ( K − ⊗ L RW ) ⊕ ( K + ⊗ h ) , (94)
which is precisely the folded complex of § 3.2 . We will next construct its L ∞ L_\infty L ∞ brackets, proceeding in two steps.
First, the ideal B + B^+ B + is acted on by A A A through the embedding tensor: an element a ∈ A a\in A a ∈ A acts on b ∈ B + b\in B^+ b ∈ B + by [ Φ ( a ) , b ] B [\Phi(a),b]_B [ Φ ( a ) , b ] B , where [ ⋅ , ⋅ ] B [\cdot,\cdot]_B [ ⋅ , ⋅ ] B denotes the bracket of the dg Lie algebra B B B , i.e. the bracket of h \mathfrak{h} h extended to forms as in (93) . Since B + B^+ B + is an ideal of B B B , this lies again in B + B^+ B + . Accordingly, the space T = A ⊕ B + T=A\oplus B^+ T = A ⊕ B + carries the brackets of a semidirect product,
μ 1 ( a + b ) = ℓ 1 A ( a ) + d B ( b ) , \mu_1(a+b)=\ell_1^A(a)+d_B(b), μ 1 ( a + b ) = ℓ 1 A ( a ) + d B ( b ) ,
μ 2 ( a 1 + b 1 , a 2 + b 2 ) = ℓ 2 A ( a 1 , a 2 ) + [ Φ a 1 , b 2 ] B + [ b 1 , Φ a 2 ] B + [ b 1 , b 2 ] B , (95) \mu_2(a_1+b_1,a_2+b_2)=\ell_2^A(a_1,a_2)+[\Phi a_1,b_2]_B+[b_1,\Phi a_2]_B+[b_1,b_2]_B, \tag*{(95)} μ 2 ( a 1 + b 1 , a 2 + b 2 ) = ℓ 2 A ( a 1 , a 2 ) + [ Φ a 1 , b 2 ] B + [ b 1 , Φ a 2 ] B + [ b 1 , b 2 ] B , (95)
μ 3 ( a 1 + b 1 , a 2 + b 2 , a 3 + b 3 ) = ℓ 3 A ( a 1 , a 2 , a 3 ) , \mu_3(a_1+b_1,a_2+b_2,a_3+b_3)=\ell_3^A(a_1,a_2,a_3), μ 3 ( a 1 + b 1 , a 2 + b 2 , a 3 + b 3 ) = ℓ 3 A ( a 1 , a 2 , a 3 ) ,
for a , a i ∈ A a,a_i\in A a , a i ∈ A and b , b i ∈ B + b,b_i\in B^+ b , b i ∈ B + , and μ n ≥ 4 = 0 \mu_{n\geq4}=0 μ n ≥ 4 = 0 . They satisfy the L ∞ L_\infty L ∞ relations: the part in A A A consists of the relations of A A A , while the part in B + B^+ B + follows from the Jacobi identity and the derivation property of d B d_B d B in B B B , because Φ \Phi Φ commutes with the differentials, intertwines the binary brackets, Φ ( ℓ 2 A ( a 1 , a 2 ) ) = [ Φ a 1 , Φ a 2 ] B \Phi(\ell_2^A(a_1,a_2))=[\Phi a_1,\Phi a_2]_B Φ ( ℓ 2 A ( a 1 , a 2 )) = [ Φ a 1 , Φ a 2 ] B , and annihilates ℓ 3 A \ell_3^A ℓ 3 A .
The above space T = A ⊕ B + T=A\oplus B^+ T = A ⊕ B + is larger than F \mathcal{F} F , since it contains the full A A A . In the second step we therefore identify the ‘high part’ of A A A with its image under the embedding tensor by dividing by the ideal
I = { a − Φ ( a ) : a ∈ A + } ⊂ T . (96) I=\{a-\Phi(a):a\in A^{+}\}\subset T . \tag*{(96)} I = { a − Φ ( a ) : a ∈ A + } ⊂ T . (96)
This is indeed an ideal of T T T , which can be seen as follows. Note first that I ⊂ T I\subset T I ⊂ T , since Φ ( A + ) ⊂ B + \Phi(A^{+})\subset B^{+} Φ ( A + ) ⊂ B + . We have to show that every bracket (95) with at least one argument in I I I lies again in I I I , i.e. is of the form c − Φ ( c ) c-\Phi(c) c − Φ ( c ) with c ∈ A + c\in A^{+} c ∈ A + . Let a ∈ A + a\in A^{+} a ∈ A + and let a ′ + b ′ a^{\prime}+b^{\prime} a ′ + b ′ , a ′ ′ + b ′ ′ a^{\prime\prime}+b^{\prime\prime} a ′′ + b ′′ be arbitrary elements of T T T . For the differential, using that Φ \Phi Φ is a chain map, d B Φ = Φ ℓ 1 A d_{B}\Phi=\Phi\ell_{1}^{A} d B Φ = Φ ℓ 1 A , we compute
μ 1 ( a − Φ ( a ) ) = ℓ 1 A ( a ) − d B Φ ( a ) = ℓ 1 A ( a ) − Φ ( ℓ 1 A ( a ) ) , (97) \mu_{1}(a-\Phi(a))=\ell_{1}^{A}(a)-d_{B}\Phi(a)=\ell_{1}^{A}(a)-\Phi\bigl(\ell_{1}^{A}(a)\bigr), \tag*{(97)} μ 1 ( a − Φ ( a )) = ℓ 1 A ( a ) − d B Φ ( a ) = ℓ 1 A ( a ) − Φ ( ℓ 1 A ( a ) ) , (97)
which is of the required form with c = ℓ 1 A ( a ) c=\ell_{1}^{A}(a) c = ℓ 1 A ( a ) . For the binary bracket, (95) with a 1 = a a_{1}=a a 1 = a , b 1 = − Φ ( a ) b_{1}=-\Phi(a) b 1 = − Φ ( a ) , a 2 = a ′ a_{2}=a^{\prime} a 2 = a ′ and b 2 = b ′ b_{2}=b^{\prime} b 2 = b ′ gives
μ 2 ( a − Φ ( a ) , a ′ + b ′ ) = ℓ 2 A ( a , a ′ ) + [ Φ a , b ′ ] B + [ − Φ a , Φ a ′ ] B + [ − Φ a , b ′ ] B = ℓ 2 A ( a , a ′ ) − [ Φ a , Φ a ′ ] B = ℓ 2 A ( a , a ′ ) − Φ ( ℓ 2 A ( a , a ′ ) ) . (98) \begin{aligned}
\mu_{2}(a-\Phi(a),a^{\prime}+b^{\prime})&=\ell_{2}^{A}(a,a^{\prime})+[\Phi a,b^{\prime}]_{B}+[-\Phi a,\Phi a^{\prime}]_{B}+[-\Phi a,b^{\prime}]_{B}\\
&=\ell_{2}^{A}(a,a^{\prime})-[\Phi a,\Phi a^{\prime}]_{B}\\
&=\ell_{2}^{A}(a,a^{\prime})-\Phi\bigl(\ell_{2}^{A}(a,a^{\prime})\bigr). \tag*{(98)}
\end{aligned} μ 2 ( a − Φ ( a ) , a ′ + b ′ ) = ℓ 2 A ( a , a ′ ) + [ Φ a , b ′ ] B + [ − Φ a , Φ a ′ ] B + [ − Φ a , b ′ ] B = ℓ 2 A ( a , a ′ ) − [ Φ a , Φ a ′ ] B = ℓ 2 A ( a , a ′ ) − Φ ( ℓ 2 A ( a , a ′ ) ) . (98)
Here, in the second step, the action [ Φ a , b ′ ] B [\Phi a,b^{\prime}]_{B} [ Φ a , b ′ ] B of a a a on b ′ b^{\prime} b ′ has cancelled against the bracket [ − Φ a , b ′ ] B [-\Phi a,b^{\prime}]_{B} [ − Φ a , b ′ ] B of the B + B^{+} B + -components, and in the last step we used that Φ \Phi Φ intertwines the binary brackets. This is of the required form with c = ℓ 2 A ( a , a ′ ) c=\ell_{2}^{A}(a,a^{\prime}) c = ℓ 2 A ( a , a ′ ) . Finally, the ternary bracket only involves the components in A A A ,
μ 3 ( a − Φ ( a ) , a ′ + b ′ , a ′ ′ + b ′ ′ ) = ℓ 3 A ( a , a ′ , a ′ ′ ) = ℓ 3 A ( a , a ′ , a ′ ′ ) − Φ ( ℓ 3 A ( a , a ′ , a ′ ′ ) ) , (99) \mu_{3}(a-\Phi(a),a^{\prime}+b^{\prime},a^{\prime\prime}+b^{\prime\prime})=\ell_{3}^{A}(a,a^{\prime},a^{\prime\prime})=\ell_{3}^{A}(a,a^{\prime},a^{\prime\prime})-\Phi\bigl(\ell_{3}^{A}(a,a^{\prime},a^{\prime\prime})\bigr), \tag*{(99)} μ 3 ( a − Φ ( a ) , a ′ + b ′ , a ′′ + b ′′ ) = ℓ 3 A ( a , a ′ , a ′′ ) = ℓ 3 A ( a , a ′ , a ′′ ) − Φ ( ℓ 3 A ( a , a ′ , a ′′ ) ) , (99)
since Φ \Phi Φ annihilates the values of ℓ 3 A \ell_{3}^{A} ℓ 3 A ; this is of the required form with c = ℓ 3 A ( a , a ′ , a ′ ′ ) c=\ell_{3}^{A}(a,a^{\prime},a^{\prime\prime}) c = ℓ 3 A ( a , a ′ , a ′′ ) . In all three cases c ∈ A + c\in A^{+} c ∈ A + , because A + A^{+} A + is an ideal of A A A .
Having established that I I I is an ideal it follows that T / I T/I T / I is an L ∞ L_{\infty} L ∞ algebra. In T / I T/I T / I every element a + b a+b a + b is equivalent to a − + ( b + Φ ( a + ) ) a^{-}+(b+\Phi(a^{+})) a − + ( b + Φ ( a + )) , which has a unique representative in A − ⊕ B + = F A^{-}\oplus B^{+}=F A − ⊕ B + = F , so that T / I ≅ F T/I\cong F T / I ≅ F . Computing the brackets (95) on such representatives and passing to the representative of the result gives the brackets of F F F : for a , a i ∈ A − a,a_{i}\in A^{-} a , a i ∈ A − and b , b i ∈ B + b,b_{i}\in B^{+} b , b i ∈ B + ,
ℓ 1 ( a + b ) = ( ℓ 1 A ( a ) ) − + Φ ( ℓ 1 A ( a ) ) + + d B ( b ) , ℓ 2 ( a 1 + b 1 , a 2 + b 2 ) = ( ℓ 2 A ( a 1 , a 2 ) ) − + Φ ( ℓ 2 A ( a 1 , a 2 ) ) + + [ Φ a 1 , b 2 ] B + [ b 1 , Φ a 2 ] B , ℓ 3 ( a 1 + b 1 , a 2 + b 2 , a 3 + b 3 ) = ( ℓ 3 A ( a 1 , a 2 , a 3 ) ) − , (100) \begin{aligned}
\ell_{1}(a+b)&=\bigl(\ell_{1}^{A}(a)\bigr)^{-}+\Phi\bigl(\ell_{1}^{A}(a)\bigr)^{+}+d_{B}(b),\\
\ell_{2}(a_{1}+b_{1},a_{2}+b_{2})&=\bigl(\ell_{2}^{A}(a_{1},a_{2})\bigr)^{-}+\Phi\bigl(\ell_{2}^{A}(a_{1},a_{2})\bigr)^{+}+[\Phi a_{1},b_{2}]_{B}+[b_{1},\Phi a_{2}]_{B},\\
\ell_{3}(a_{1}+b_{1},a_{2}+b_{2},a_{3}+b_{3})&=\bigl(\ell_{3}^{A}(a_{1},a_{2},a_{3})\bigr)^{-}, \tag*{(100)}
\end{aligned} ℓ 1 ( a + b ) ℓ 2 ( a 1 + b 1 , a 2 + b 2 ) ℓ 3 ( a 1 + b 1 , a 2 + b 2 , a 3 + b 3 ) = ( ℓ 1 A ( a ) ) − + Φ ( ℓ 1 A ( a ) ) + + d B ( b ) , = ( ℓ 2 A ( a 1 , a 2 ) ) − + Φ ( ℓ 2 A ( a 1 , a 2 ) ) + + [ Φ a 1 , b 2 ] B + [ b 1 , Φ a 2 ] B , = ( ℓ 3 A ( a 1 , a 2 , a 3 ) ) − , (100)
and ℓ n ≥ 4 = 0 \ell_{n\geq4}=0 ℓ n ≥ 4 = 0 . Here [ b 1 , b 2 ] B [b_{1},b_{2}]_{B} [ b 1 , b 2 ] B no longer appears, since the antiholomorphic degrees of two elements of the right half add up to at least four.
These arguments prove the following proposition.
Proposition 8: F F F with the brackets (100) is an L ∞ L_{\infty} L ∞ algebra. Its relations follow from the facts that K \mathcal{K} K is a graded commutative differential graded algebra with differential ideal K + \mathcal{K}^{+} K + , that L R W L_{\mathrm{RW}} L RW is an L ∞ L_{\infty} L ∞ algebra, that H \mathcal{H} H is a dg Lie algebra, and that Π \Pi Π is a strict morphism.
Let us spell out these brackets. The differential is ∂ \partial ∂ and ∂ ˉ \bar{\partial} ∂ ˉ within both halves, while the bridge is obtained by first applying ∂ ˉ \bar{\partial} ∂ ˉ and then the embedding tensor ∂ \partial ∂ , which reproduces (3.21).
The brackets of two elements of the left half that stay in the left half are the brackets of A A A , i.e. those of the Roytenberg–Weinstein algebra with the antiholomorphic forms multiplied,
ℓ 2 ( λ 1 , λ 2 ) = [ λ 1 , λ 2 ] C , ℓ 2 ( λ , e ) = [ λ , e ] C , ℓ 2 ( λ , χ ) = 1 2 L ϑ ( λ ) ∂ χ , \ell_{2}(\lambda_{1},\lambda_{2})=[\lambda_{1},\lambda_{2}]_{C},\qquad\ell_{2}(\lambda,e)=[\lambda,e]_{C},\qquad\ell_{2}(\lambda,\chi)=\frac{1}{2}\mathcal{L}_{\vartheta(\lambda)}\partial\chi, ℓ 2 ( λ 1 , λ 2 ) = [ λ 1 , λ 2 ] C , ℓ 2 ( λ , e ) = [ λ , e ] C , ℓ 2 ( λ , χ ) = 2 1 L ϑ ( λ ) ∂ χ ,
ℓ 2 ( λ , λ ˉ ) = 1 2 L ϑ ( λ ) λ ˉ , ℓ 2 ( χ , e ) = − 1 2 ι ϑ ( e ) ∂ χ . (101) \ell_{2}(\lambda,\bar{\lambda})=\frac{1}{2}\mathcal{L}_{\vartheta(\lambda)}\bar{\lambda},\qquad\ell_{2}(\chi,e)=-\frac{1}{2}\iota_{\vartheta(e)}\partial\chi. \tag*{(101)} ℓ 2 ( λ , λ ˉ ) = 2 1 L ϑ ( λ ) λ ˉ , ℓ 2 ( χ , e ) = − 2 1 ι ϑ ( e ) ∂ χ . (101)
The only bracket of two elements of the left half that is carried across is
ℓ 2 ( e 1 , e 2 ) = Φ ( ℓ 2 A ( e 1 , e 2 ) ) = ∂ [ e 1 , e 2 ] C = [ ∂ e 1 , ∂ e 2 ] B , (102) \ell_{2}(e_{1},e_{2})=\Phi\bigl(\ell_{2}^{A}(e_{1},e_{2})\bigr)=\partial[e_{1},e_{2}]_{C}=[\partial e_{1},\partial e_{2}]_{B}, \tag*{(102)} ℓ 2 ( e 1 , e 2 ) = Φ ( ℓ 2 A ( e 1 , e 2 ) ) = ∂ [ e 1 , e 2 ] C = [ ∂ e 1 , ∂ e 2 ] B , (102)
where the last step is the strictness of Π \Pi Π . The brackets of an element of the left half with one of the right half are
ℓ 2 ( λ , y ∗ ) = [ ∂ λ , y ∗ ] B = L ϑ ( λ ) y ∗ , ℓ 2 ( e , e ∗ ) = [ ∂ e , e ∗ ] B , ℓ 2 ( e , λ ˉ ∗ ) = [ ∂ e , λ ˉ ∗ ] B , (103) \ell_{2}(\lambda,y^{*})=[\partial\lambda,y^{*}]_{B}=\mathcal{L}_{\vartheta(\lambda)}y^{*},\qquad\ell_{2}(e,e^{*})=[\partial e,e^{*}]_{B},\qquad\ell_{2}(e,\bar{\lambda}^{*})=[\partial e,\bar{\lambda}^{*}]_{B}, \tag*{(103)} ℓ 2 ( λ , y ∗ ) = [ ∂ λ , y ∗ ] B = L ϑ ( λ ) y ∗ , ℓ 2 ( e , e ∗ ) = [ ∂ e , e ∗ ] B , ℓ 2 ( e , λ ˉ ∗ ) = [ ∂ e , λ ˉ ∗ ] B , (103)
with the bracket [ ⋅ , ⋅ ] B [\cdot,\cdot]_{B} [ ⋅ , ⋅ ] B of B B B . Finally, the ternary brackets are those of A A A that stay in the left half,
ℓ 3 ( λ 1 , λ 2 , λ 3 ) = − T ( λ 1 , λ 2 , λ 3 ) , ℓ 3 ( λ 1 , λ 2 , e ) = − T ( λ 1 , λ 2 , e ) . (104) \ell_{3}(\lambda_{1},\lambda_{2},\lambda_{3})=-T(\lambda_{1},\lambda_{2},\lambda_{3}),\qquad\ell_{3}(\lambda_{1},\lambda_{2},e)=-T(\lambda_{1},\lambda_{2},e). \tag*{(104)} ℓ 3 ( λ 1 , λ 2 , λ 3 ) = − T ( λ 1 , λ 2 , λ 3 ) , ℓ 3 ( λ 1 , λ 2 , e ) = − T ( λ 1 , λ 2 , e ) . (104)
With (4.5) and ∂ α = ι ϑ ( α ) ω \partial\alpha=\iota_{\vartheta(\alpha)}\omega ∂ α = ι ϑ ( α ) ω , the brackets (4.15) and (4.16) are those of (3.23), the bracket of H H H on Ω 2 , ∙ \Omega^{2,\bullet} Ω 2 , ∙ being the Lie bracket of vector-valued forms transported by ω \omega ω , i.e. minus the Schouten bracket. Hence F F F coincides with the L ∞ L_{\infty} L ∞ algebra L L L of section 3.2, and Proposition 8 proves Proposition 4. Together with Proposition 5 this also proves the master equation for (3.35).
Including cohomology and moduli
So far we have ignored global issues. The prepotential (3.1) solves the constraint ∂ A ′ = 0 \partial A'=0 ∂ A ′ = 0 precisely if every ∂ \partial ∂ -closed ( 2 , 1 ) (2,1) ( 2 , 1 ) -form is ∂ \partial ∂ -exact. The obstruction to this is the cohomology H 2 , 1 H^{2,1} H 2 , 1 , which vanishes on C 3 \mathbb{C}^{3} C 3 and some other manifolds of interest for twisted holography [8 ] . On a compact Calabi–Yau threefold, Hodge theory implies that every ∂ \partial ∂ -closed ( 2 , 1 ) (2,1) ( 2 , 1 ) -form is the sum of a ∂ \partial ∂ -exact form and a unique harmonic form t t t , where the latter cannot be written as ∂ e \partial e ∂ e [41 ] . We include these harmonic ( 2 , 1 ) (2,1) ( 2 , 1 ) -forms, which describe the infinitesimal deformations of the complex structure (complex structure moduli) [46 , 47 , 48 ] , by enlarging the folded complex by the harmonic ( 2 , 1 ) (2,1) ( 2 , 1 ) -forms and their duals,
L e x t = L ⊕ H 2 , 1 ⊕ H 1 , 2 , (105) L^{\mathrm{ext}}=L\oplus\mathcal{H}^{2,1}\oplus\mathcal{H}^{1,2}, \tag*{(105)} L ext = L ⊕ H 2 , 1 ⊕ H 1 , 2 , (105)
with t ∈ H 2 , 1 t\in\mathcal{H}^{2,1} t ∈ H 2 , 1 in degree 1 and t ∗ ∈ H 1 , 2 t^{*}\in\mathcal{H}^{1,2} t ∗ ∈ H 1 , 2 in degree 2, extending the differential by zero and the inner product by ⟨ ⟨ t , t ∗ ⟩ ⟩ = ∫ X t ∗ ∧ t \langle\langle t,t^{*}\rangle\rangle=\int_{X}t^{*}\wedge t ⟨⟨ t , t ∗ ⟩⟩ = ∫ X t ∗ ∧ t . The full Beltrami differential is simply shifted as
A = ϑ ( e ) + t ^ , (106) A=\vartheta(e)+\hat{t}, \tag*{(106)} A = ϑ ( e ) + t ^ , (106)
where we note that also t ^ \hat{t} t ^ is divergence-free, since harmonic forms on a compact Kähler manifold are d d d -closed, so that ∂ t = 0 \partial t=0 ∂ t = 0 , and (2.14) gives ( div ω t ^ ) ⌟ ω = − ∂ t = 0 (\operatorname{div}_{\omega}\hat{t})\mathbin{\lrcorner}\omega=-\partial t=0 ( div ω t ^ ) ┘ ω = − ∂ t = 0 . Equivalently, A ′ = ∂ e + t A'=\partial e+t A ′ = ∂ e + t satisfies the constraint ∂ A ′ = 0 \partial A'=0 ∂ A ′ = 0 .9
The brackets ℓ 2 ( e , e ) \ell_2(e,e) ℓ 2 ( e , e ) , ℓ 2 ( e , e ∗ ) \ell_2(e,e^*) ℓ 2 ( e , e ∗ ) and ℓ 2 ( e , λ ˉ ∗ ) \ell_2(e,\bar{\lambda}^*) ℓ 2 ( e , λ ˉ ∗ ) of L L L depend on e e e only through ϑ ( e ) \vartheta(e) ϑ ( e ) , and the substitution ϑ ( e ) → ϑ ( e ) + t ^ \vartheta(e)\to\vartheta(e)+\hat{t} ϑ ( e ) → ϑ ( e ) + t ^ gives the new brackets
ℓ 2 ( e , t ) = − { ϑ ( e ) , t ^ } ⌟ ω , ℓ 2 ( t 1 , t 2 ) = − { t ^ 1 , t ^ 2 } ⌟ ω , \ell_2(e,t)=-\{\vartheta(e),\hat{t}\}\lrcorner\omega,\qquad\ell_2(t_1,t_2)=-\{\hat{t}_1,\hat{t}_2\}\lrcorner\omega, ℓ 2 ( e , t ) = − { ϑ ( e ) , t ^ } ┘ ω , ℓ 2 ( t 1 , t 2 ) = − { t ^ 1 , t ^ 2 } ┘ ω ,
ℓ 2 ( t , e ∗ ) = { t ^ , e ∗ ⌟ ω − 1 } ⌟ ω , ℓ 2 ( t , λ ˉ ∗ ) = ∂ ( t ^ ∧ c ) ⌟ ω . (107) \ell_2(t,e^*)=\{\hat{t},e^*\lrcorner\omega^{-1}\}\lrcorner\omega,\qquad\ell_2(t,\bar{\lambda}^*)=\partial(\hat{t}\wedge c)\lrcorner\omega. \tag*{(107)} ℓ 2 ( t , e ∗ ) = { t ^ , e ∗ ┘ ω − 1 } ┘ ω , ℓ 2 ( t , λ ˉ ∗ ) = ∂ ( t ^ ∧ c ) ┘ ω . (107)
The remaining brackets of L L L involving e e e would require a prepotential for t t t , which does not exist globally. They are replaced by
ℓ 2 ( λ , t ) = ι ϑ ( λ ) t − ∂ ι t ^ λ , ℓ 2 ( χ , t ) = − ι t ^ ∂ χ . (108) \ell_2(\lambda,t)=\iota_{\vartheta(\lambda)}t-\partial\iota_{\hat{t}}\lambda,\qquad\ell_2(\chi,t)=-\iota_{\hat{t}}\partial\chi. \tag*{(108)} ℓ 2 ( λ , t ) = ι ϑ ( λ ) t − ∂ ι t ^ λ , ℓ 2 ( χ , t ) = − ι t ^ ∂ χ . (108)
while no ternary bracket involving t t t is needed, and all brackets with an argument t ∗ t^* t ∗ vanish.
Proposition 9: The brackets of L L L together with (107) and (108) satisfy all L ∞ L_\infty L ∞ relations with values in L L L .
This follows from the double copy as before, with the left half enlarged by H 2 , 1 \mathcal{H}^{2,1} H 2 , 1 : the harmonic forms lie in B − B^- B − , but not in the image of Φ \Phi Φ , and one extends Φ \Phi Φ by Φ ( t ) = t \Phi(t)=t Φ ( t ) = t . In particular, ∂ ℓ 2 ( λ , t ) = ∂ ι ϑ ( λ ) t = L ϑ ( λ ) t = [ ∂ λ , t ] \partial\ell_2(\lambda,t)=\partial\iota_{\vartheta(\lambda)}t=\mathcal{L}_{\vartheta(\lambda)}t=[\partial\lambda,t] ∂ ℓ 2 ( λ , t ) = ∂ ι ϑ ( λ ) t = L ϑ ( λ ) t = [ ∂ λ , t ] by Cartan’s formula and ∂ t = 0 \partial t=0 ∂ t = 0 , while ℓ 2 ( χ , t ) \ell_2(\chi,t) ℓ 2 ( χ , t ) lies in the kernel of Φ \Phi Φ . We have also verified the relations explicitly, using that locally every harmonic ( 2 , 1 ) (2,1) ( 2 , 1 ) -form is of the form ∂ ∂ ˉ β ˉ \partial\bar{\partial}\bar{\beta} ∂ ∂ ˉ β ˉ . The brackets with values in H 1 , 2 \mathcal{H}^{1,2} H 1 , 2 are fixed by cyclicity.
The field equation becomes ∂ ∂ ˉ e − 1 2 { A , A } ⌟ ω = 0 \partial\bar{\partial}e-\frac{1}{2}\{A,A\}\lrcorner\omega=0 ∂ ∂ ˉ e − 2 1 { A , A } ┘ ω = 0 , i.e. the Kodaira–Spencer equation for the full Beltrami differential (4.19), and the gauge transformations become δ e = ∂ λ ˉ + ∂ ˉ λ − [ λ , e ] C − ι ϑ ( λ ) t + ∂ ι t ^ λ \delta e=\partial\bar{\lambda}+\bar{\partial}\lambda-[\lambda,e]_C-\iota_{\vartheta(\lambda)}t+\partial\iota_{\hat{t}}\lambda δ e = ∂ λ ˉ + ∂ ˉ λ − [ λ , e ] C − ι ϑ ( λ ) t + ∂ ι t ^ λ . The isomorphism of Proposition 5 extends to the moduli without any further components. Although the original L ∞ L_\infty L ∞ algebra has no ternary brackets involving t t t , the isomorphism generates the ternary bracket ℓ 3 c y c ( λ 1 , λ 2 , t ) = Q ( λ 1 , λ 2 , t ) \ell_3^{\mathrm{cyc}}(\lambda_1,\lambda_2,t)=Q(\lambda_1,\lambda_2,t) ℓ 3 cyc ( λ 1 , λ 2 , t ) = Q ( λ 1 , λ 2 , t ) , where
Q ( λ 1 , λ 2 , t ) = ι ϑ ( λ 1 ) ι ϑ ( λ 2 ) t − L ϑ ( λ 1 ) ι t ^ λ 2 + L ϑ ( λ 2 ) ι t ^ λ 1 + ι t ^ [ λ 1 , λ 2 ] C . (109) Q(\lambda_1,\lambda_2,t)=\iota_{\vartheta(\lambda_1)}\iota_{\vartheta(\lambda_2)}t-\mathcal{L}_{\vartheta(\lambda_1)}\iota_{\hat{t}}\lambda_2+\mathcal{L}_{\vartheta(\lambda_2)}\iota_{\hat{t}}\lambda_1+\iota_{\hat{t}}[\lambda_1,\lambda_2]_C. \tag*{(109)} Q ( λ 1 , λ 2 , t ) = ι ϑ ( λ 1 ) ι ϑ ( λ 2 ) t − L ϑ ( λ 1 ) ι t ^ λ 2 + L ϑ ( λ 2 ) ι t ^ λ 1 + ι t ^ [ λ 1 , λ 2 ] C . (109)
The cyclic algebra is then encoded in the master action
S e x t = ∫ X [ 1 2 e ∧ ∂ ∂ ˉ e − e ∗ ∧ ( ∂ ˉ λ + ∂ λ ˉ − L ϑ ( λ ) e − ι ϑ ( λ ) t + ∂ ι t ^ λ ) + λ ∗ ∧ ( ∂ χ + 1 2 [ λ , λ ] C ) + χ ∗ ∧ ( 1 2 ι ϑ ( λ ) ∂ χ + 1 6 T ( λ , λ , λ ) ) + λ ˉ ∗ ∧ ( ∂ ˉ χ − ι t ^ ∂ χ + L ϑ ( λ ) λ ˉ + 1 2 ℓ 2 ( λ , λ ) ∣ Ω 0 , 1 + 1 2 Q ( λ , λ , t ) ) ] + 1 6 ∫ X ω ∧ ( ( A ∧ A ∧ A ) ⌟ ω ) . (110) \begin{aligned}
S_{\mathrm{ext}}={}&\int_X\Bigl[\frac{1}{2}e\wedge\partial\bar{\partial}e-e^*\wedge\bigl(\bar{\partial}\lambda+\partial\bar{\lambda}-\mathcal{L}_{\vartheta(\lambda)}e-\iota_{\vartheta(\lambda)}t+\partial\iota_{\hat{t}}\lambda\bigr)+\lambda^*\wedge\bigl(\partial\chi+\frac{1}{2}[\lambda,\lambda]_C\bigr)\\
&+\chi^*\wedge\bigl(\frac{1}{2}\iota_{\vartheta(\lambda)}\partial\chi+\frac{1}{6}T(\lambda,\lambda,\lambda)\bigr)+\bar{\lambda}^*\wedge\bigl(\bar{\partial}\chi-\iota_{\hat{t}}\partial\chi+\mathcal{L}_{\vartheta(\lambda)}\bar{\lambda}\\
&\qquad+\frac{1}{2}\ell_2(\lambda,\lambda)\big|_{\Omega^{0,1}}+\frac{1}{2}Q(\lambda,\lambda,t)\bigr)\Bigr]+\frac{1}{6}\int_X\omega\wedge\bigl((A\wedge A\wedge A)\lrcorner\omega\bigr). \tag*{(110)}
\end{aligned} S ext = ∫ X [ 2 1 e ∧ ∂ ∂ ˉ e − e ∗ ∧ ( ∂ ˉ λ + ∂ λ ˉ − L ϑ ( λ ) e − ι ϑ ( λ ) t + ∂ ι t ^ λ ) + λ ∗ ∧ ( ∂ χ + 2 1 [ λ , λ ] C ) + χ ∗ ∧ ( 2 1 ι ϑ ( λ ) ∂ χ + 6 1 T ( λ , λ , λ ) ) + λ ˉ ∗ ∧ ( ∂ ˉ χ − ι t ^ ∂ χ + L ϑ ( λ ) λ ˉ + 2 1 ℓ 2 ( λ , λ ) Ω 0 , 1 + 2 1 Q ( λ , λ , t ) ) ] + 6 1 ∫ X ω ∧ ( ( A ∧ A ∧ A ) ┘ ω ) . (110)
with A A A as in (106) . For t = 0 t=0 t = 0 this reduces to (70) . The cubic term now contains the Yukawa coupling 1 6 C a b c t a t b t c \frac{1}{6}C_{abc}t^at^bt^c 6 1 C ab c t a t b t c , with C a b c = ∫ X ω ∧ ( ( t ^ a ∧ t ^ b ∧ t ^ c ) ⌟ ω ) C_{abc}=\int_X\omega\wedge((\hat{t}_a\wedge\hat{t}_b\wedge\hat{t}_c)\lrcorner\omega) C ab c = ∫ X ω ∧ (( t ^ a ∧ t ^ b ∧ t ^ c ) ┘ ω ) .
The moduli enter (110) without kinetic term and without gauge transformation, and t ∗ t^* t ∗ does not appear, so that S e x t S_{\mathrm{ext}} S ext is best viewed as a family of BV theories for the local fields, parametrized by t t t . Varying t t t is not part of Kodaira–Spencer theory: by the Tian–Todorov lemma [46 , 47 ] , which in our conventions follows directly from (21) as { β , γ } = ( − 1 ) β div ω ( β ∧ γ ) \{\beta,\gamma\}=(-1)^\beta\operatorname{div}_{\omega}(\beta\wedge\gamma) { β , γ } = ( − 1 ) β div ω ( β ∧ γ ) for div ω β = div ω γ = 0 \operatorname{div}_{\omega}\beta=\operatorname{div}_{\omega}\gamma=0 div ω β = div ω γ = 0 , the bracket { A , A } \{A,A\} { A , A } has no harmonic part, so that the harmonic part of the Kodaira–Spencer equation vanishes identically and the moduli are unobstructed. They are thus background data, as in [3 ] .
Conclusions and outlook
In this paper we have presented a local action principle for Kodaira–Spencer gravity. To this end the constraint ∂ A ′ = 0 \partial A^{\prime}=0 ∂ A ′ = 0 , on which the non-local action of BCOV is based [2 , 3 ] , is solved in terms of a ( 1 , 1 ) (1,1) ( 1 , 1 ) -form e e e by writing A ′ = ∂ e A^{\prime}=\partial e A ′ = ∂ e . This construction is a natural generalization of [27 ] , which gave a local reformulation of Bandos–Townsend theories [49 ] . Remarkably, apart from making the action local, this induces at the level of the free theory precisely the gauge transformations of double field theory. Moreover, the non-linear gauge structure is based on a Courant algebroid and its associated L ∞ L_{\infty} L ∞ algebra, precisely as in standard double field theory. In addition, the L ∞ L_{\infty} L ∞ algebra defining the full theory has a double copy interpretation that is analogous to [32 , 33 ] . The present theory differs substantially from standard double field theory, however, in that its classical action only has cubic couplings, and no level-matching or section constraint is needed for consistency.
The present formulation clarifies in which sense Kodaira–Spencer theory is really a theory of gravity (albeit a topological one). Concretely, we truncated the theory to the three-dimensional real slice z = z ˉ z=\bar{z} z = z ˉ for which the gauge transformations reduce to the spin-2 and B-field gauge transformations, while the quadratic action reduces to the sum of Fierz–Pauli action for massless spin-2 and the action for a Kalb–Ramond B-field. At the non-linear level the theory differs, however, from ordinary double field theory or Einstein–Hilbert gravity coupled to a B-field. Indeed, the theory is exact with only cubic couplings, which carry three derivatives and violate parity, in contrast to the parity even and non-polynomial two-derivative interactions of Einstein–Hilbert theory. To cubic order, and truncating the antisymmetric B-field, the action coincides with an exotic 3D gravity coupling found by Boulanger and Gualtieri [31 ] . Integrating out the B-field leads to non-localities that, however, can be removed by non-local field redefinitions, at least up to and including quartic order.
Kodaira–Spencer gravity has also been related to three-dimensional Einstein gravity in recent work. In particular, Erdmenger et al. studied Kodaira–Spencer gravity on S L ( 2 , C ) ≅ H 3 × S 3 SL(2,\mathbb{C})\cong H^{3}\times S^{3} S L ( 2 , C ) ≅ H 3 × S 3 [38 , 50 ] , a background relevant for twisted holography [8 ] . They showed that, upon restricting to a subsector of complex structure deformations defined by a ‘locality constraint’ stronger than ∂ A ′ = 0 \partial A^{\prime}=0 ∂ A ′ = 0 , the prepotential can be expressed algebraically in terms of the Beltrami differential, up to ∂ ˉ \bar{\partial} ∂ ˉ -closed terms. This leads to a holomorphic Chern–Simons action for S L ( 2 , C ) SL(2,\mathbb{C}) S L ( 2 , C ) , whose dimensional reduction yields chiral A d S 3 AdS_{3} A d S 3 gravity. It would be interesting to investigate this relation further using the present formulation. (The cohomological obstruction discussed in § 4.3 vanishes on this background [51 ] .)
Furthermore, Ben-Shahar et al. analyzed a double copy of 3D Chern–Simons theory and showed that it is on-shell equivalent to 3D gravity in a unimodular gauge [39 ] , see also [52 ] , with the volume-preserving diffeomorphisms as gauge symmetries. In terms of a divergence-free field A a μ A^{\mu}_{a} A a μ , the action has a non-local quadratic term and a cubic term, while a local formulation in terms of constrained vector frames is also available. This is closely related to the real version of the non-local action (32) ; identifying the holomorphic vector index of the Beltrami differential with μ \mu μ and its antiholomorphic form index with a a a , and correspondingly z ˉ \bar{z} z ˉ with z z z , one recovers the double copy action of [53 ] , generalized in [37 ] . Our prepotential then provides a local parametrization of the divergence-free field, as seen in the reduction of § 3.4 .
The present work should be extended in various directions. Most intriguingly, it may provide a powerful toy model to study how gravity may be completed to a theory in higher dimensions with an improved UV behavior. Indeed, the 3D gravity theory found here through dimensional reduction is badly non-renormalizable by power counting (although it has no local degrees of freedom), yet its higher-dimensional 6D origin in Kodaira–Spencer gravity may have a well-defined quantization. Indeed, Costello and Li developed the perturbative quantization of BCOV theory in the BV formalism [5 , 6 ] . It would be interesting to study such features in the present formulation, which is more ‘gravity-like’, and to see whether it may shed light on possible quantizations of non-topological gravity. Finally, it would be interesting to study twisted holography in the present formulation.
Acknowledgments and use of AI
The main idea of this paper is mine, but everything else is heavily based on interactions with artificial intelligence, particularly with Claude Opus 5.5, which I used in all aspects: for the brainstorming of ideas, literature search, explicit computations, and the write-up of results. For instance, the double copy interpretation of § 4 is essentially due to Claude, although the final presentation went through countless iterations, transforming an initially highly involved mathematical proposal to a reasonably simple one. Moreover, any AI-written parts of the text went through many iterations to hopefully eliminate all instances of ‘AI slop’. After completing a first final draft I used ChatGPT (GPT-6) for a critical review of the entire manuscript and in particular to independently derive and check the L ∞ L_{\infty} L ∞ brackets transported by the isomorphism (66) , including the absence of higher brackets, and to verify the cyclicity condition (64) . I take responsibility for the entire paper.
Moreover, I would like to thank my human colleagues Roberto Bonezzi, Nicolas Boulanger, Christoph Chiaffrino and Barton Zwiebach for discussions and comments on this project, and Kevin Costello for alerting me to ref. [9 ] .