Spectral sequence of an isometric action Thanks: This work has been partially supported by Ministerio de Ciencia e Innovación, Spain, grant PID2022-139631NB-I00. The authors acknowledge that the research cooperation was funded by the program Excellence Initiative Research University at the Jagiellonian University in Krakow within the framework of the research group Reeb-Reinhardt 2022. José Ignacio Royo Prieto, Martintxo Saralegi-Aranguren Abstract . We consider a free smooth action Φ : G × M → M \Phi\colon G\times M\to M of a connected compact Lie group G G on a manifold M M . We examine the Cartan filtration of the complex of differential forms of M M . The associated spectral sequence E p , q r {E}^{{}^{p,q}}_{{}_{r}} converges to the cohomology of M M . It is well known that the second page E p , q 2 {E}^{{}^{p,q}}_{{}_{2}} of this spectral sequence is given by H p ( M / G ) ⊗ H q ( 𝔤 ) {H}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)} , where 𝔤 \noindent{\mathfrak{g}} denotes the Lie algebra of G G . In this note, we provide a straightforward proof of this fact without using Mayer-Vietoris, harmonic operators, or other such methods found in existing proofs. In fact, we extend this result to the case where the action is locally free and G G is not compact, under the hypothesis that Φ \Phi extends to a smooth action of a compact Lie group K K . The compactness of K K is a crucial aspect of our proof. When G G is not compact, the cohomology H p ( M / G ) {H}^{{}^{p}}{\left(M/G\right)} is not the cohomology of the orbit space M / G M/G , which may be a topologically wild space, but rather the basic cohomology of the foliation determined by the action of G G . 2010 Mathematics Subject Classification Primary 57S15; Secondary 55N33. 1. Introduction We consider in this note a locally free smooth action \Phi\colon G\times M\to M of a connected Lie group G on a manifold M . We also assume the existence of a smooth extension of \Phi to the action \Psi\colon K\times M\to M of a compact Lie group K . The Cartan filtration classifies the differential forms of M based on the number of vector fields tangent to the orbits of \Phi required to annihilate them. This filtration defines a spectral sequence, the Leray-Serre-deRham spectral sequence {E}^{{}^{p,q}}_{{}_{r}} , which converges to the cohomology of M . The main result of this note is the isomorphism (1) {E}^{{}^{p,q}}_{{}_{2}}\cong{H}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}, where {H}^{{}^{p}}{\left(M/G\right)} denotes the basic cohomology of the foliation determined by the orbits of the action \Phi , and \noindent{\mathfrak{g}} is the Lie algebra of G . This result has already been proven by other authors [ 2 , 1 , 7 ] using different techniques such as Mayer-Vietoris on M/G , harmonic analysis on M , etc. The compactness of K is the keystone of our approach. First of all, we prove that, thanks to the compactness of K , we can restrict ourselves to using invariant differential forms to compute {E}^{{}^{p,q}}_{{}_{2}} (see Proposition 6.4 ). In other words, we begin with {\underline{E}}^{{}^{p,q}}_{{}_{0}}\cong\left({\Omega}^{{}^{p}}_{{}_{hor}}{\left(M\right)}\otimes{\bigwedge}^{{}^{q}}(\raise 2.0pt\hbox{$\chi$}_{1},\ldots,\raise 2.0pt\hbox{$\chi$}_{n})\right)^{G}. We then carry out an explicit computation of the second term of the spectral sequence by simply using the classical decomposition of the differential d=d_{2,-1}+d_{1,0}+d_{0,1} associated to a Riemannian metric of M . A priori, the operator d_{0,1} mixes the tangent bundle of the action with its orthogonal bundle (see ( 18 )). Once again, the compactness of K allows us to construct a rich Riemannian metric, which we call a good metric (see Proposition 3.3 ), for which d_{0,1} lives in the tangent bundle of the action (see Proposition 5.2 ). This gives {\underline{E}}^{{}^{p,q}}_{{}_{1}}\cong\left({\Omega}^{{}^{p}}_{{}_{hor}}{\left(M\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}\right)^{G}={\Omega}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)} (cf. ( 25 )). The computation of {E}^{{}^{p,q}}_{{}_{2}} is thus achieved and we get ( 1 ) (see Proposition 6.4 and Theorem 7.2 ). 2. The group G We summarize here all the tools coming from a Lie group G , connected subgroup of a compact Lie group K , that we will use in this work, including the fundamental vector fields and the associated characteristic forms. Let G be a connected subgroup of a compact Lie group K . We consider a bi-invariant Riemannian metric \nu on G , which exists because K is compact. The Lie algebra of G is denoted by \noindent{\mathfrak{g}} . The associated fundamental vector field , relative to the right action of G on itself 1 1 1 We could have interchanged left with right with the same result. , for a given u\in\noindent{\mathfrak{g}} , is denoted by Y_{u} . These vector fields verify: \displaystyle(L_{g})_{*}Y_{u}= \displaystyle Y_{u},\hbox{ and} \displaystyle(R_{g})_{*}Y_{u}= \displaystyle Y_{\mathop{\rm Ad\,}\nolimits(g^{-1})\cdot u} for any u\in\noindent{\mathfrak{g}} and g\in G . Here • L_{g}\colon G\to G is defined by L_{g}(\ell)=g\cdot\ell , • R_{g}\colon G\to G is defined by R_{g}(\ell)=\ell\cdot g^{-1} , and • \mathop{\rm Ad\,}\nolimits\colon G\to\mathop{\rm Aut\,}\nolimits(\noindent{\mathfrak{g}}) is the adjoint representation of G . Notice that the vector fields Y_{\bullet} are G -left invariant vector fields. Given u\in\noindent{\mathfrak{g}} we write \gamma_{u}=i_{Y_{u}}\nu\in{\Omega}^{{}^{1}}{\left(G\right)} the associated characteristic form associated with u . They verify (2) \displaystyle L_{g}^{*}\gamma_{u}= \displaystyle\gamma_{u},\hbox{ and } \displaystyle R_{g}^{*}\gamma_{u}= \displaystyle\gamma_{\mathop{\rm Ad\,}\nolimits(g^{-1})\cdot u} for any u\in\noindent{\mathfrak{g}} and g\in G . Notice that these differential forms \gamma_{\bullet} are G -left invariant. Identifying the Lie algebra \noindent{\mathfrak{g}} with the Lie algebra generated by the family \{X_{1}\ldots,X_{n}\} , the complex \bigwedge^{*}\noindent{\mathfrak{g}}^{*} is in fact isomorphic to the complex of left-invariant differential forms of G . We have the isomorphism (3) {H}^{{}^{*}}{\left(\noindent{\mathfrak{g}}\right)}\cong\left(\bigwedge^{*}\noindent{\mathfrak{g}}^{*}\right)^{G} where (\bigwedge^{*}\noindent{\mathfrak{g}}^{*})^{G} is the complex of G-invariant elements, where we have considered the right action of G (see, for example, [ 4 , Chap. 4, Sec.2] ). Notice that \left(\bigwedge^{*}\noindent{\mathfrak{g}}^{*}\right)^{G} is then identified (and is isomorphic as a dgca) with the complex of bi-invariant diferential forms on G. Consider the metric \langle-,-\rangle defined on the Lie algebra \noindent{\mathfrak{g}} by \langle u,v\rangle=\nu(Y_{u},Y_{v}) . We fix for the sequel an orthonormal basis \{u_{1},\ldots,u_{n}\} for \noindent{\mathfrak{g}} . For simplicity, we denote Y_{u_{i}}=Y_{i}, and \gamma_{u_{i}}=\gamma_{i}, for each i\in\{1,\ldots n\} . Notice that \gamma_{i}(Y_{j})=\delta_{i,j} for any i,j\in\{1,\ldots n\} . This metric \langle-,-\rangle is \mathop{\rm Ad\,}\nolimits -invariant since for each g\in G and each u,v\in\noindent{\mathfrak{g}} , we have: (4) \displaystyle\langle\mathop{\rm Ad\,}\nolimits(g^{-1})\cdot u,\mathop{\rm Ad\,}\nolimits(g^{-1})\cdot v\rangle=\nu(Y_{\mathop{\rm Ad\,}\nolimits(g^{-1})\cdot u},Y_{\mathop{\rm Ad\,}\nolimits(g^{-1})\cdot v}\rangle=\nu((R_{g^{-1}})_{*}Y_{u},(R_{g^{-1}})_{*}Y_{v}\rangle=\nu(Y_{u},Y_{v})=\langle{u},{v}\rangle, since \nu is bi-invariant. Proposition 2.1 . We have (5) d\gamma_{u}=-\sum_{1\leq a0 . Consider the differential form \eta=\omega-\sum_{1\leq i_{1}<\cdotsq , where \omega=\eta\wedge\raise 2.0pt\hbox{$\chi$}_{i_{1}}\wedge\cdots\wedge\raise 2.0pt\hbox{$\chi$}_{i_{q}} and \eta is a horizontal form. By using ( 15 ) and ( 6 ) we finally have \displaystyle\rho(\omega)(X_{j_{0}},\ldots,X_{j_{r}}) \displaystyle=\int_{K}a^{*}\omega(X_{j_{0}},\ldots,X_{j_{r}})da=\int_{K}\left(a^{*}\eta\wedge a^{*}\raise 2.0pt\hbox{$\chi$}_{i_{1}}\wedge\cdots\wedge a^{*}\raise 2.0pt\hbox{$\chi$}_{i_{q}}\right)(X_{j_{0}},\ldots,X_{j_{r}})da \displaystyle=\pm\int_{K}a^{*}\eta\wedge a^{*}(\raise 2.0pt\hbox{$\chi$}_{i_{1}}\wedge\cdots\wedge\raise 2.0pt\hbox{$\chi$}_{i_{q}})(X_{j_{0}},\ldots,X_{j_{r}})da=\pm\int_{K}a^{*}\eta\wedge(\raise 2.0pt\hbox{$\chi$}_{i_{1}}\wedge\cdots\wedge\raise 2.0pt\hbox{$\chi$}_{i_{q}})(a_{*}X_{j_{0}},\ldots,a_{*}X_{j_{r}})da=0. iv) The first property is immediate from the definition of \rho . Any k\in K preserves the complexes {\Omega}^{{}^{*}}_{{}_{hor}}{\left(M\right)} and {\bigwedge}^{{}^{*}}(\raise 2.0pt\hbox{$\chi$}_{1},\ldots,\raise 2.0pt\hbox{$\chi$}_{n}) (cf. ( 15 ) and ( 7 )). Hence, if \omega\in{\underline{\Omega}}^{{}^{p,q}}{\left(M\right)} , then \rho(\omega)\in{\underline{\Omega}}^{{}^{p,q}}{\left(M\right)} . Since d\rho(\omega)=\rho(d\omega) we obtain (1\otimes\delta)\rho(\omega)=\rho((1\otimes\delta)\omega) for degree reasons. The last assertion follows similarly from g^{*}d=dg^{*} . ∎ Proposition 6.4 . The relationship between the two LSdR spectral sequences we have introduced is {\underline{E}}^{{}^{*,*}}_{{}_{r}}{\left(M\right)}\cong{E}^{{}^{*,*}}_{{}_{r}}{\left(M\right)}.\hfill for each r\geq 2 . Proof. This follows from Lemma 6.3 i), ii) and iii)(cf. [ 6 , Exercice 20, pag.78] ). ∎ 7. Computation of the second term We provide the computation of the second term of the LSdR spectral sequence. Before beginning this computation, we need two remarks about the notation. A) The cohomology {H}^{{}^{*}}{\left(\noindent{\mathfrak{g}}\right)} is isomorphic to the space of invariant forms \left(\bigwedge^{*}\noindent{\mathfrak{g}}^{*}\right)^{G} (cf. ( 3 )). Notice that the two following dgcas are isomorphic: \left(\bigwedge^{*}\noindent{\mathfrak{g}}^{*},d\right)=\left(\bigwedge^{*}\left(\gamma_{1},\ldots,\gamma_{n}\right),d\right)\ \ \hbox{ and }\ \ \ \left(\bigwedge^{*}\left(\chi_{1},\ldots,\chi_{n}\right),d\right) via the assignment \gamma_{u}\mapsto\chi_{u} (cf. ( 5 ) and ( 12 )). Moreover, the Lie group G acts on both dgcas preserving the aforementioned assignment (cf. ( 2 ) and ( 7 )). Consequently, \left(\bigwedge^{*}\left(\chi_{1},\ldots,\chi_{n}\right)\right)^{G} is isomorphic to the cohomology {H}^{{}^{*}}{\left(\noindent{\mathfrak{g}}\right)} . B) The first term of the invariant LSdR spectral sequence is {\underline{E}}^{{}^{p,q}}_{{}_{1}}=\frac{{\underline{Z}}^{{}^{p,q}}_{{}_{1}}(M)}{{\underline{B}}^{{}^{p,q}}_{{}_{0}}(M)+{\underline{Z}}^{{}^{p+1,q-1}}_{{}_{0}}(M)}, with \displaystyle{\underline{Z}}^{{}^{p,q}}_{{}_{1}}(M) \displaystyle= \displaystyle\left\{\eta\in{F}^{{}^{p}}{\underline{\Omega}}^{{}^{p+q}}{\left(M\right)}\mid d\eta\in{F}^{{}^{p+1}}{\underline{\Omega}}^{{}^{p+q+1}}{\left(M\right)}\right\}={F}^{{}^{p+1}}{\underline{\Omega}}^{{}^{p+q}}{\left(M\right)}\oplus\left\{\eta\in{\underline{\Omega}}^{{}^{p,q}}{\left(M\right)}\mid({\large 1}\otimes\delta)\ \eta=0\right\}, (24) \displaystyle{\underline{B}}^{{}^{p,q}}_{{}_{0}}(M) \displaystyle= \displaystyle\left\{d\eta\in{F}^{{}^{p}}{\underline{\Omega}}^{{}^{p+q}}{\left(M\right)}\mid\eta\in{F}^{{}^{p}}{\underline{\Omega}}^{{}^{p+q-1}}{\left(M\right)}\right\}=d{F}^{{}^{p}}{\underline{\Omega}}^{{}^{p+q-1}}{\left(M\right)},\hbox{ and} \displaystyle{\underline{Z}}^{{}^{p+1,q-1}}_{{}_{0}}(M) \displaystyle= \displaystyle\left\{\eta\in{F}^{{}^{p+1}}{\underline{\Omega}}^{{}^{p+q}}{\left(M\right)}\mid d\eta\in{F}^{{}^{p+1}}{\underline{\Omega}}^{{}^{p+q+1}}{\left(M\right)}\right\}={F}^{{}^{p+1}}{\underline{\Omega}}^{{}^{p+q}}{\left(M\right)}. Lemma 7.1 . Let \eta\in{\underline{\Omega}}^{{}^{p,q}}{\left(M\right)} with ({\large 1}\otimes\delta)\eta=0. Then, there exists \eta^{\prime}\in{\underline{\Omega}}^{{}^{p,q-1}}{\left(M\right)} with \eta-({\large 1}\otimes\delta)\eta^{\prime}\in{\Omega}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}. Proof. Consider the cohomology class [\eta]\in{H}^{{}^{q}}{\left({\Omega}^{{}^{p}}_{{}_{hor}}{\left(M\right)}\otimes\bigwedge^{*}\left(\chi_{1},\ldots,\chi_{n}\right),{\large 1}\otimes\delta\right)}={\Omega}^{{}^{p}}_{{}_{hor}}{\left(M\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)} . Notice that (25) [\eta]\in\left({\Omega}^{{}^{p}}_{{}_{hor}}{\left(M\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}\right)^{G}={\Omega}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}={\Omega}^{{}^{p}}{\left(M/G\right)}\otimes\left(\bigwedge^{q}\left(\chi_{1},\ldots,\chi_{n}\right)\right)^{G} (cf. ( 3 )). This gives the decomposition \eta=\eta_{0}+({\large 1}\otimes\delta)\eta^{\prime\prime}, where \eta_{0}\in{\Omega}^{{}^{p}}{\left(M/G\right)}\otimes\left(\bigwedge^{q}\left(\chi_{1},\ldots,\chi_{n}\right)\right)^{G} and \eta^{\prime\prime}\in{\Omega}^{{}^{p,q-1}}{\left(M\right)} . Applying the operator \rho of Lemma 6.3 iv) we get \eta=\rho(\eta)=\rho(\eta_{0})+\rho({\large 1}\otimes\delta)\eta^{\prime\prime}=\eta_{0}+\underbrace{({\large 1}\otimes\delta)\rho(\eta^{\prime\prime})}_{\eta^{\prime}} with \eta^{\prime}\in{\underline{\Omega}}^{{}^{p,q-1}}{\left(M\right)} . ∎ We have arrived at the main result of this work. Theorem 7.2 . Let \Phi\colon G\times M\to M be a locally free tame action and let {E}^{{}^{p,q}}_{{}_{2}} be the second term of the associated LSdR spectral sequence. Then, the map F\colon{H}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}\to{E}^{{}^{p,q}}_{{}_{2}} given by 3 3 3 For the calculations, we use {H}^{{}^{*}}{\left(\noindent{\mathfrak{g}}\right)}=\left(\bigwedge^{*}(\chi_{1},\ldots,\chi_{n})\right)^{G} , as explained in Section 2 . F([\alpha]\otimes\beta)=\mathop{\rm class\,}\nolimits(\alpha\otimes\beta). is an isomorphism, where {H}^{{}^{p}}{\left(M/G\right)} denotes the basic cohomology of the foliation defined by the action, and {H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)} is the cohomology of the Lie algebra \noindent{\mathfrak{g}} of G . Proof. The result comes from a commutative diagram \textstyle{{\Omega}^{{}^{p}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\cong} \scriptstyle{F} \scriptstyle{d\otimes{\large 1}} \textstyle{{E}^{{}^{p,q}}_{{}_{1}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{d_{1}} \textstyle{{\Omega}^{{}^{p+1}}{\left(M/G\right)}\otimes{H}^{{}^{q}}{\left(\noindent{\mathfrak{g}}\right)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces} \scriptstyle{\cong} \scriptstyle{F} \textstyle{{E}^{{}^{p+1,q}}_{{}_{1}}.} where d_{1} becomes d_{1}(\mathop{\rm class\,}\nolimits(\alpha\wedge\beta))=\mathop{\rm class\,}\nolimits(d\alpha\wedge\beta) (cf. ( 7 ) and Proposition 5.2 ). Notice that {E}^{{}^{p,q}}_{{}_{1}}=\frac{\left\{\eta\in{\underline{\Omega}}^{{}^{p,q}}{\left(M\right)}\mid({\large 1}\otimes\delta)\ \eta=0\right\}}{\left\{({\large 1}\otimes\delta)\eta\in{\underline{\Omega}}^{{}^{p,q}}{\left(M\right)}\mid\ \eta\in{\underline{\Omega}}^{{}^{p,q-1}}{\left(M\right)}\right\}}. The operator F is defined as follows: F(\alpha\otimes\beta)=\mathop{\rm class\,}\nolimits(\alpha\wedge\beta). It is clearly a well-defined monomorphism and satisfies d_{1}F=F(d\otimes{\large 1}) . It is an epimorphism by Lemma 7.1 . ∎