Real involutive systems on compact Lie groups Thanks: Supported by the São Paulo Research Foundation (FAPESP, grants 2023/11769-5, 2024/08416-6, 2024/12753-8, 2025/08642-9), by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, grants 163837/2022-8, 170244/2023-7, 404175/2023-6, 303483/2024-5, 308433/2025-4, 308705/2025-4), by Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG, grant APQ-00020-24), and by Deutsche Forschungsgemeinschaft (DFG, grant JA 3453/2-1). Gabriel Araújo, Igor A. Ferra, Max R. Jahnke, Luis F. Ragognette Abstract . On a compact connected Lie group G G , we study the global solvability and the cohomology spaces of the differential complex associated with an essentially real involutive structure that is invariant under left translations. We prove that solvability in the first degree of the complex implies solvability in all other degrees, and furnish a converse for this fact under a certain commutativity hypothesis (that always holds when G G is a torus). Additionally, it is proved that the solvability holds when the structure comes from the Lie algebra of a closed subgroup of G G . We also investigate real tube structures when G G is the base manifold. 2020 Mathematics Subject Classification 35F35 (primary), 58J10, 35R03 (secondary) 1. Introduction Let \Omega be a smooth manifold and \mathcal{V} an involutive subbundle of \mathbb{C}T\Omega , the complexified tangent bundle of \Omega . Involutivity means that the Lie bracket of any two sections of \mathcal{V} (which are, in particular, complex vector fields over \Omega ) is again a section of \mathcal{V} . Such a situation arises in several instances of interest, but particularly in complex geometry: if \Omega is a complex manifold, and \mathcal{V} is spanned by the antiholomorphic vector fields; or if \Omega is a real hypersurface in a complex manifold, and \mathcal{V} is spanned by the antiholomorphic vector fields which are tangent to \Omega . The latter situation is the object of study of CR geometry, and is concerned with phenomena such as extendability of holomorphic functions across the hypersurface \Omega , thus playing a key role in the function theory of the ambient manifold. Another situation of this kind, more intimately related to our own, arises in the theory of foliations. In PDE terminology, we will assume that \mathcal{V} is essentially real , meaning simply that \mathcal{V} is the complexification of an involutive subbundle of T\Omega ; i.e., an integrable distribution on \Omega . That distribution, in turn, produces, by means of the Frobenius Theorem, a foliation on \Omega whose leaves are tangent to \mathcal{V} . In either case, to such an involutive bundle \mathcal{V} , a general abstract construction (see [ 8 , 18 ] ) gives rise to a complex of vector bundles over \Omega and first-order differential operators between them (1.1) {\lx@inpgf@ignorespaces\cdots} {\lx@inpgf@ignorespaces\mathscr{C}^{\infty}(\Omega;\underline{\Lambda}^{q-1})} {\lx@inpgf@ignorespaces\mathscr{C}^{\infty}(\Omega;\underline{\Lambda}^{q})} {\lx@inpgf@ignorespaces\mathscr{C}^{\infty}(\Omega;\underline{\Lambda}^{q+1})} {\lx@inpgf@ignorespaces\cdots} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{d}^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{d}^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{d}^{\prime}} \scriptstyle{\lx@inpgf@ignorespaces\mathrm{d}^{\prime}} which encodes important information of the underlying geometry. In each of the three cases discussed above, such a construction yields: the Dolbeault complex, in complex geometry; the \bar{\partial}_{b} complex, in CR geometry; and the foliated de Rham complex, in foliation theory. In the present paper, our objects of interest are the cohomology spaces H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(\Omega)) of the complex ( 1.1 ) when \mathcal{V} is essentially real and \Omega is compact. Except when \mathcal{V}=\mathbb{C}T\Omega – in which case ( 1.1 ) is just the de Rham complex –, the complex ( 1.1 ) is not elliptic. Therefore, one cannot expect H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(\Omega)) to be finite dimensional, so we compromise on a more modest goal. Notice that all the spaces in ( 1.1 ) carry natural Fréchet topologies, for which the differential operators \mathrm{d}^{\prime} are continuous, hence the H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(\Omega)) inherit a quotient topology, making them topological vector spaces: our task will be to determine necessary or sufficient conditions for H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(\Omega)) to be Hausdorff. From the point of view of the global analysis of PDEs, the latter property is central, and is called global solvability in degree q : by Functional Analysis, it is nothing but closedness of the range of \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(\Omega;\underline{\Lambda}^{q-1})\to\mathscr{C}^{\infty}(\Omega;\underline{\Lambda}^{q}) . At this level of generality, however, such questions are very hard to answer, so we turn our attention to special cases involving symmetries. Left-invariant structures on compact Lie groups The case to which we dedicated most of our work is when the manifold is a Lie group and \mathcal{V} is invariant under left translations (hence called a left-invariant subbundle ). In this context, let G be a compact, connected Lie group and \mathcal{V}\subset\mathbb{C}TG be an essentially real, left-invariant subbundle. Our first main result is the following: Theorem 1 . If (1.2) \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G)\longrightarrow\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1})\quad\text{has closed range} then \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q-1})\longrightarrow\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})\quad\text{has closed range} for all 1\leq q\leq n\doteq\rank(\mathcal{V}) . Moreover, in that case, we have (1.3) H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))\cong\mathscr{K}(G)\otimes H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G}),\quad 0\leq q\leq n, topologically, where: • \mathscr{K}(G) is the space of homogeneous solutions of \mathcal{V} : an f\in\mathscr{C}^{\infty}(G) belongs to \mathscr{K}(G) if and only if \mathrm{L}f=0 whenever \mathrm{L} is a smooth section of \mathcal{V} ; and • H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G}) is the space of left-invariant classes 2 in H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G)) . Results regarding global solvability at intermediate levels ( q\geq 2 ) are scarce compared to the first level of the complex. Our result was possible thanks to a decomposition at the level of the cohomology groups done in Section 4 . With this decomposition in hands, our result is a consequence of Theorem 5.1 and Remark 4.2 . Properly interpreted, the spirit of Theorem 1 can be traced back to [ 7 , 11 ] , where a single (Diophantine) condition simultaneously governs solvability of all degrees of the complex for certain involutive structures on tori called tube structures (more on that later on). These two works were the major conceptual inspiration for our results: on a product of tori, by using a normalization process (Section 7.1 ), they reduce the study of a tube structure to a left-invariant one. Recently, a similar approach was carried out in [ 12 ] for the product of torus and a compact Lie group. Here, that “single (Diophantine) condition” is encoded in ( 1.2 ), which always holds for two special, disjoint classes of structures: (1) When \mathcal{V} is globally hypoelliptic , that is: any distribution f\in\mathscr{D}^{\prime}(G) such that \mathrm{L}f\in\mathscr{C}^{\infty}(G) for every \mathrm{L} section of \mathcal{V} is already smooth to start with. This is the same as global hypoellipticity of the differential operator \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G)\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1}) , which then has closed range as a consequence of [ 3 , Proposition 3.11] . In the globally hypoelliptic case, we also show (Corollary 5.3 ) that formula ( 1.3 ) simplifies further, since \mathscr{K}(G) only contains constant functions, hence H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))\cong H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G}),\quad 0\leq q\leq n. (2) When the leaves of the foliation generated by \mathcal{V} are closed in G (Theorem 6.1 ). Notice that, since \mathcal{V} is left-invariant, such leaves are just translates of H\subset G , the integral subgroup associated with the Lie algebra \mathfrak{v} of left-invariant sections of \mathcal{V} . In this case, formula ( 1.3 ) becomes (Corollary 6.2 ): H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))\cong\mathscr{C}^{\infty}(G/H)\otimes H^{q}_{\mathrm{dR}}(H),\quad 0\leq q\leq n. Our Theorem 1 is clearly an extension of (and, incidentally, provide alternative proof for) a classical result in [ 9 ] asserting the existence of left-invariant representatives of de Rham cohomology classes on G . That work was the third major influence in the present paper, and inspired the results in the Appendix. It is also very much related to a deep question posed by one of the authors [ 15 ] , in the context of elliptic structures, concerning the existence of such left-invariant representatives in various settings. When G is a torus, the converse of Theorem 1 is known [ 7 ] . In our setting, Theorem 5.4 provides a partial converse under the additional assumption that the Lie algebra \mathfrak{v} of left-invariant sections of \mathcal{V} is Abelian. This condition, however, does not preclude the non-commutativity of G itself. Tube structures Let M be a compact, connected manifold, and \mathbb{T}^{m}\doteq\mathbb{R}^{m}/2\pi\mathbb{Z}^{m} be the m -dimensional torus. There is a particular way to induce an involutive structure on \Omega\doteq M\times\mathbb{T}^{m} , called a (corank m ) tube structure , which we now describe. Let \omega_{1},\ldots,\omega_{m} be smooth, closed 1 -forms on M , and take \mathcal{V}\subset\mathbb{C}T(M\times\mathbb{T}^{m}) the subbundle whose sections are vector fields annihilated by all the 1 -forms \zeta_{k}\doteq\mathrm{d}x_{k}-\omega_{k},\quad k\in\{1,\ldots,m\}; here, x=(x_{1},\ldots,x_{m}) are the usual coordinates on \mathbb{T}^{m} . The global solvability of the differential complex ( 1.1 ) associated with a tube structure is a question of great interest (see for instance [ 2 , 5 , 6 , 13 , 14 ] or [ 10 ] for non-compact manifolds). If M is a Lie group (with Lie algebra \mathfrak{m} ) then so is G\doteq M\times\mathbb{T}^{m} (with Lie algebra \mathfrak{g}=\mathfrak{m}\oplus\mathbb{R}^{m} ), and \mathcal{V} is essentially real and left-invariant if and only if \omega_{1},\ldots,\omega_{m} are real-valued and left-invariant. While this is still not quite in the context of Theorem 1 , we prove in Corollary 7.3 that every real tube structure on G is equivalent to a left-invariant one by means of a diffeomorphism. Since the differential complex ( 1.1 ) is an invariant functorially attached to \mathcal{V} , by means of this “normalization” we can replace the original tube structure by a much simpler one (of “constant coefficients”, in a certain sense), and take full advantage of the symmetries of G to analyze it: for now the whole machinery of Theorem 1 is available. We exploit this to investigate a question raised in a previous paper [ 2 ] . There, while looking for a certain representation formula for H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(\Omega)) for tube structures (akin to the one presented in [ 11 ] when M is a torus), we found an obstruction, which vanishes in certain cases (e.g., the torus), but not always: we then conjectured that the obstruction should vanish when M is a Lie group (hence, the said representation formula does hold). In the present paper, our last main result is the proof of this conjecture (Corollary 7.5 ). These questions are explained in details in Section 7.2 , and settled in Section 7.3 via perturbations of the de Rham complex by closed 1 -forms. 2. Preliminaries We assume that G is compact and connected Lie group and recall that an involutive subbundle \mathcal{V}\subset\mathbb{C}TG is left-invariant if (\ell_{x})_{*}\mathcal{V}_{y}\subset\mathcal{V}_{xy},\quad\forall x,y\in G; here, \ell_{x}:G\to G stands for the left translation by x\in G . Such bundles are in one-to-one correspondence with complex Lie subalgebras \mathfrak{v} of \mathbb{C}\mathfrak{g} – the Lie algebra of all complex, left-invariant vector fields on G – in a natural way. Here, we are interested in the case when \mathcal{V} is essentially real 1 1 1 This is not necessary in this section, though. , meaning that \mathcal{V}_{x}=\bar{\mathcal{V}}_{x} for every x\in G which is equivalent to say that \mathfrak{v}=\bar{\mathfrak{v}} as subspaces of \mathbb{C}\mathfrak{g} . The latter means that we can choose \mathrm{L}_{1},\ldots,\mathrm{L}_{n} real left-invariant vector fields forming a basis for \mathfrak{v} . We choose additional vectors \mathrm{M}_{1},\ldots,\mathrm{M}_{m}\in\mathbb{C}\mathfrak{g} such that, together, they form a basis for \mathbb{C}\mathfrak{g} . We denote the dual of \mathbb{C}\mathfrak{g} by \mathbb{C}\mathfrak{g}^{*} and the dual basis of \mathrm{L}_{1},\ldots,\mathrm{L}_{n},\mathrm{M}_{1},\ldots,\mathrm{M}_{m}\in\mathbb{C}\mathfrak{g} by \tau_{1},\ldots,\tau_{n},\zeta_{1},\ldots,\zeta_{m}\in\mathbb{C}\mathfrak{g}^{*} . These are left-invariant 1 -forms on G , and, since m+n=\dim G , for f\in\mathscr{C}^{\infty}(G) we can write \mathrm{d}f=\sum_{j=1}^{n}(\mathrm{L}_{j}f)\ \tau_{j}+\sum_{k=1}^{m}(\mathrm{M}_{k}f)\ \zeta_{k}. For q\in\{0,\ldots,n\} , let \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) be the space of smooth q -forms on G of type (2.1) u=\sideset{}{{}^{\prime}}{\sum}_{|J|=q}u_{J}\ \tau_{J},\quad u_{J}\in\mathscr{C}^{\infty}(G). The prime indicates that the summation is restricted to ordered multi-indices J=(j_{1},\ldots,j_{q}) , where we take \tau_{J}\doteq\tau_{j_{1}}\wedge\cdots\wedge\tau_{j_{q}} . The exterior derivative induces a differential operator \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})\longrightarrow\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q+1}) as follows: given u\in\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) as in ( 2.1 ) we have \mathrm{d}u=\sum_{j=1}^{n}\sideset{}{{}^{\prime}}{\sum}_{|J|=q}(\mathrm{L}_{j}u_{J})\ \tau_{j}\wedge\tau_{J}+\sum_{k=1}^{m}\sideset{}{{}^{\prime}}{\sum}_{|J|=q}(\mathrm{M}_{k}u_{J})\ \zeta_{k}\wedge\tau_{J}+\sideset{}{{}^{\prime}}{\sum}_{|J|=q}u_{J}\ \mathrm{d}\tau_{J}, which we project back onto \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q+1}) by deleting any term containing a \zeta_{k} ; for instance, the whole second double sum. As for the third sum, since \mathrm{d}\tau_{J} is a left-invariant (q+1) -form on G , it can be written as a linear combination over \mathbb{C} of our basic forms, and we only care about the portions of such a combination which happen to live in \underline{\Lambda}^{q+1} . That is, there are constants \alpha_{JK}\in\mathbb{C} such that (2.2) \mathrm{d}\tau_{J}=\sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}\alpha_{JK}\ \tau_{K}+\text{terms containing at least one $\zeta_{k}$}, leading us to (2.3) \mathrm{d}^{\prime}u\doteq\sum_{j=1}^{n}\sideset{}{{}^{\prime}}{\sum}_{|J|=q}(\mathrm{L}_{j}u_{J})\ \tau_{j}\wedge\tau_{J}+\sideset{}{{}^{\prime}}{\sum}_{|J|=q}\sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}\alpha_{JK}u_{J}\ \tau_{K}. We have that \mathrm{d}^{\prime}\circ\mathrm{d}^{\prime}=0 , i.e., a differential complex is defined: our purpose is to investigate the cohomology spaces H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))\doteq\frac{\ker\{\mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})\longrightarrow\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q+1})\}}{\ran\{\mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q-1})\longrightarrow\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})\}},\quad q\in\{0,\ldots,n\}. A useful (and more invariant) point of view is to regard a q -form w on G as an alternating \mathscr{C}^{\infty}(G) -multilinear map w:\mathscr{C}^{\infty}(G;\mathbb{C}TG)^{q}\to\mathscr{C}^{\infty}(G) . In that sense, we define \mathscr{C}^{\infty}(G;\T^{\prime 1,q-1}) as the space of all such w satisfying w(\mathrm{X}_{1},\ldots,\mathrm{X}_{q})=0\quad\text{whenever $\mathrm{X}_{1},\ldots,\mathrm{X}_{q}$ are sections of $\mathcal{V}$}. One checks at once that every equivalence class in \mathscr{C}^{\infty}(G;\wedge^{q}\mathbb{C}T^{*}G)/\mathscr{C}^{\infty}(G;\T^{\prime 1,q-1}) has a unique representative in \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) ; that these two spaces are therefore isomorphic; and that \mathrm{d}^{\prime} is, in this regard, precisely the map induced on the quotients by the exterior derivative. Actually, in our concrete setup \mathscr{C}^{\infty}(G;\T^{\prime 1,q-1}) is exactly our set of remainders (“terms containing at least one \zeta_{k} ”), hence \mathscr{C}^{\infty}(G;\wedge^{q}\mathbb{C}T^{*}G)=\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})\oplus\mathscr{C}^{\infty}(G;\T^{\prime 1,q-1}). Thanks to the invariant formula for the exterior derivative [ 17 , Proposition 12.19] , for any u\in\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) the following holds: if \mathrm{X}_{0},\mathrm{X}_{1},\ldots,\mathrm{X}_{q} are sections of \mathcal{V} then (2.4) (\mathrm{d}^{\prime}u)(\mathrm{X}_{0},\ldots,\mathrm{X}_{q})=\sum_{k=0}^{q}(-1)^{k}\mathrm{X}_{k}\left(u(\mathrm{X}_{0},\ldots,\widehat{\mathrm{X}}_{k},\ldots,\mathrm{X}_{q})\right)\\ +\sum_{j0 there exists C_{s}>0 such that (3.4) \|\mathcal{F}_{\lambda}(f)\|_{L^{2}(G)}\leq C_{s}(1+\lambda)^{-s},\quad\forall\lambda\in\sigma(\Delta_{G}), in which case the series ( 3.3 ) actually converges in \mathscr{C}^{\infty}(G) . Since every \mathrm{X}\in\mathfrak{g} is skew-symmetric with respect to the L^{2} inner product on functions, one easily shows that \Delta_{G}=\Delta_{\mathfrak{g}} (for any choice of orthonormal basis of \mathfrak{g} ), hence \Delta_{G} commutes with any \mathrm{X}\in\mathfrak{g} (Lemma 2.1 ), and then also with \Delta_{\mathfrak{v}} . In particular, \Delta_{\mathfrak{v}}(E_{\lambda})\subset E_{\lambda} for every \lambda\in\sigma(\Delta_{G}) : we define \mathscr{K}_{\lambda}\doteq\ker\left\{\Delta_{\mathfrak{v}}:E_{\lambda}\longrightarrow E_{\lambda}\right\},\quad\mathscr{R}_{\lambda}\doteq\ran\left\{\Delta_{\mathfrak{v}}:E_{\lambda}\longrightarrow E_{\lambda}\right\}. From ( 3.2 ) we further infer that E_{\lambda}=\mathscr{K}_{\lambda}\oplus\mathscr{R}_{\lambda} as a sum of orthogonal, finite dimensional subspaces. Given f\in\mathscr{C}^{\infty}(G) , using the L^{2} -orthogonality we are entitled to decompose uniquely \mathcal{F}_{\lambda}(f)=\mathcal{F}_{\lambda}(f)_{\mathscr{K}}+\mathcal{F}_{\lambda}(f)_{\mathscr{R}} in \mathscr{K}_{\lambda}\oplus\mathscr{R}_{\lambda} , for each \lambda\in\sigma(\Delta_{G}) . It follows from the Paley-Wiener Theorem above that both series f_{\mathscr{K}}\doteq\sum_{\lambda\in\sigma(\Delta_{G})}\mathcal{F}_{\lambda}(f)_{\mathscr{K}},\quad f_{\bar{\mathscr{R}}}\doteq\sum_{\lambda\in\sigma(\Delta_{G})}\mathcal{F}_{\lambda}(f)_{\mathscr{R}} converge in \mathscr{C}^{\infty}(G) . Clearly, f=f_{\mathscr{K}}+f_{\bar{\mathscr{R}}} and f_{\mathscr{K}}\perp f_{\bar{\mathscr{R}}} . On the one hand, it is clear that \Delta_{\mathfrak{v}}(f_{\mathscr{K}})=\Delta_{\mathfrak{v}}\Big(\sum_{\lambda\in\sigma(\Delta_{G})}\mathcal{F}_{\lambda}(f)_{\mathscr{K}}\Big)=\sum_{\lambda\in\sigma(\Delta_{G})}\Delta_{\mathfrak{v}}\mathcal{F}_{\lambda}(f)_{\mathscr{K}}=0 i.e., f_{\mathscr{K}}\in\mathscr{K}(G) . On the other hand, by definition of \mathscr{R}_{\lambda} , we can solve \Delta_{\mathfrak{v}}u_{\lambda}=\mathcal{F}_{\lambda}(f)_{\mathscr{R}},\quad u_{\lambda}\in E_{\lambda}, for each \lambda\in\sigma(\Delta_{G}) , hence f_{\bar{\mathscr{R}}}=\sum_{\lambda\in\sigma(\Delta_{G})}\mathcal{F}_{\lambda}(f)_{\mathscr{R}}=\lim_{\nu\to\infty}\sum_{\begin{subarray}{c}\lambda\in\sigma(\Delta_{G})\\ |\lambda|\leq\nu\end{subarray}}\Delta_{\mathfrak{v}}u_{\lambda} with convergence in \mathscr{C}^{\infty}(G) : we conclude that f_{\bar{\mathscr{R}}}\in\bar{\mathscr{R}}(G) , since as each truncated sum above belongs to \mathscr{R}(G) . Finally, to check continuity of the projections \pi_{\mathscr{K}},\pi_{\bar{\mathscr{R}}} (we endow \mathscr{K}(G),\bar{\mathscr{R}}(G) with their subspace topologies inherited from \mathscr{C}^{\infty}(G) , which turn them into Fréchet spaces), we make use of adapted Sobolev norms: for f\in\mathscr{C}^{\infty}(G) and k\in\mathbb{Z}_{+} , let \|f\|_{\mathscr{H}^{k}(G)}\doteq\|(I+\Delta_{G})^{k}f\|_{L^{2}(G)}=\Big(\sum_{\lambda\in\sigma(\Delta_{G})}(1+\lambda)^{2k}\|\mathcal{F}_{\lambda}(f)\|_{L^{2}(G)}^{2}\Big)^{\frac{1}{2}}; they generate the topology of \mathscr{C}^{\infty}(G) and, given \mathscr{F}\in\{\mathscr{K},\bar{\mathscr{R}}\} , it is clear that \|\pi_{\mathscr{F}}(f)\|_{\mathscr{H}^{k}(G)}^{2}\leq\|\pi_{\mathscr{K}}(f)\|_{\mathscr{H}^{k}(G)}^{2}+\|\pi_{\bar{\mathscr{R}}}(f)\|_{\mathscr{H}^{k}(G)}^{2}=\|f\|_{\mathscr{H}^{k}(G)}^{2}, from which continuity of \pi_{\mathscr{F}} is immediate. ∎ We extend the previous result to forms: Corollary 3.2 . We have \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})=\mathscr{K}(G;\underline{\Lambda}^{q})\oplus\bar{\mathscr{R}}(G;\underline{\Lambda}^{q}) as a sum of closed subspaces. The associated projections \pi_{\mathscr{K}},\pi_{\bar{\mathscr{R}}} are continuous, and commute with \mathrm{d}^{\prime} . Proof. It is clear that \mathscr{K}(G;\underline{\Lambda}^{q})\cap\bar{\mathscr{R}}(G;\underline{\Lambda}^{q})=\{0\} due to their coefficientwise definition. We write u\in\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) as u=\sideset{}{{}^{\prime}}{\sum}_{|J|=q}u_{J}\ \tau_{J}=\sideset{}{{}^{\prime}}{\sum}_{|J|=q}\left(\pi_{\mathscr{K}}(u_{J})+\pi_{\bar{\mathscr{R}}}(u_{J})\right)\ \tau_{J}, hence u=\pi_{\mathscr{K}}(u)+\pi_{\bar{\mathscr{R}}}(u) , where \pi_{\mathscr{F}}(u)\doteq\sideset{}{{}^{\prime}}{\sum}_{|J|=q}\pi_{\mathscr{F}}(u_{J})\tau_{J}\in\mathscr{F}(G;\underline{\Lambda}^{q}),\quad\mathscr{F}\in\{\mathscr{K},\bar{\mathscr{R}}\}. All the claims are now obvious. For the last one, notice that \mathrm{d}^{\prime}u=\mathrm{d}^{\prime}\pi_{\mathscr{K}}(u)+\mathrm{d}^{\prime}\pi_{\bar{\mathscr{R}}}(u) , where \mathrm{d}^{\prime}\pi_{\mathscr{F}}(u)\in\mathscr{F}(G;\underline{\Lambda}^{q+1}),\quad\mathscr{F}\in\{\mathscr{K},\bar{\mathscr{R}}\}, as both \mathscr{K}(G),\bar{\mathscr{R}}(G) are \mathfrak{v} -invariant spaces. Uniqueness in the direct sum decomposition \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q+1})=\mathscr{K}(G;\underline{\Lambda}^{q+1})\oplus\bar{\mathscr{R}}(G;\underline{\Lambda}^{q+1}) shows that \mathrm{d}^{\prime}\pi_{\mathscr{F}}(u)=\pi_{\mathscr{F}}(\mathrm{d}^{\prime}u) . ∎ 4. Decompositions in cohomology Given two \mathfrak{v} -invariant subspaces \mathscr{F}(G)\subset\mathscr{F}^{\bullet}(G)\subset\mathscr{C}^{\infty}(G) , the continuous inclusion maps \mathscr{F}(G;\underline{\Lambda}^{q})\hookrightarrow\mathscr{F}^{\bullet}(G;\underline{\Lambda}^{q}) define chain maps, hence induce linear maps in cohomology (recall their definition in ( 2.7 )) H^{q}_{\mathrm{d}^{\prime}}(\mathscr{F}(G))\longrightarrow H^{q}_{\mathrm{d}^{\prime}}(\mathscr{F}^{\bullet}(G)),\quad q\in\{0,\ldots,n\}, continuous for the quotient topologies; they may be neither injective nor surjective. For \mathscr{F}\in\{\mathscr{K},\bar{\mathscr{R}}\} , we constructed in the previous section projections \pi_{\mathscr{F}}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})\to\mathscr{F}(G;\underline{\Lambda}^{q}) , which are also continuous chain maps, and hence induce continuous linear maps in cohomology: H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))\longrightarrow H^{q}_{\mathrm{d}^{\prime}}(\mathscr{F}(G)),\quad q\in\{0,\ldots,n\}. Since the composition {\lx@inpgf@ignorespaces\mathscr{F}(G;\underline{\Lambda}^{q})} {\lx@inpgf@ignorespaces\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q})} {\lx@inpgf@ignorespaces\mathscr{F}(G;\underline{\Lambda}^{q})} \scriptstyle{\lx@inpgf@ignorespaces\pi_{\mathscr{F}}} is the identity map, so is the induced composition (4.1) {\lx@inpgf@ignorespaces H^{q}_{\mathrm{d}^{\prime}}(\mathscr{F}(G))} {\lx@inpgf@ignorespaces H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))} {\lx@inpgf@ignorespaces H^{q}_{\mathrm{d}^{\prime}}(\mathscr{F}(G)),} thus making the first (resp. second) arrow in ( 4.1 ) injective (resp. surjective). In this sense, we regard both H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G)) , H^{q}_{\mathrm{d}^{\prime}}(\bar{\mathscr{R}}(G)) as linear subspaces of H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G)) : algebraically, we have the decomposition (4.2) H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G))\cong H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G))\oplus H^{q}_{\mathrm{d}^{\prime}}(\bar{\mathscr{R}}(G)) whose factors we now investigate. Consider 2 2 2 Here, \mathbb{C}_{G} stands for the space of constant functions on G , and \mathbb{C}_{G}(G;\underline{\Lambda}^{q}) for the (finite dimensional) space of left-invariant forms u in \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) ; i.e., those whose coefficients u_{J} in representation ( 2.1 ) are all constant . the linear map \mathsf{T}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\to\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1}) defined by \mathsf{T}\bigg(\sideset{}{{}^{\prime}}{\sum}_{|J|=q}c_{J}\ \tau_{J}\bigg)\doteq\sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}\bigg(\sideset{}{{}^{\prime}}{\sum}_{|J|=q}\alpha_{JK}c_{J}\bigg)\ \tau_{K} where \alpha_{JK}\in\mathbb{C} are given by ( 2.2 ), i.e.: (4.3) \mathrm{d}^{\prime}\tau_{J}=\sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}\alpha_{JK}\ \tau_{K}. By standard arguments of Linear Algebra, one can find a linear map \mathsf{S}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})\longrightarrow\mathbb{C}_{G}(G;\underline{\Lambda}^{q}) such that \mathsf{T}\circ\mathsf{S} is the identity on \ran\mathsf{T} . Proposition 4.1 . The range of \mathrm{d}^{\prime}:\mathscr{K}(G;\underline{\Lambda}^{q})\to\mathscr{K}(G;\underline{\Lambda}^{q+1}) is always closed. That is: H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G)) is Hausdorff (hence, a Fréchet space) for each q\in\{0,\ldots,n\} . Proof. By means of ( 2.1 ), we interpret elements in \mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) as smooth functions on G valued on the finite dimensional vector space \mathbb{C}_{G}(G;\underline{\Lambda}^{q}) . In that sense, we note that if u\in\mathscr{K}(G;\underline{\Lambda}^{q}) then for every x\in G have (4.4) (\mathrm{d}^{\prime}u)(x)=\sideset{}{{}^{\prime}}{\sum}_{|J|=q}\sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}\alpha_{JK}u_{J}(x)\ \tau_{K}=\mathsf{T}(u(x)). Now let \{u_{\nu}\}_{\nu\in\mathbb{N}}\subset\mathscr{K}(G;\underline{\Lambda}^{q}) be such that \mathrm{d}^{\prime}u_{\nu}\to f\in\mathscr{K}(G;\underline{\Lambda}^{q+1}) . By ( 2.3 ): (4.5) \sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}\bigg(\sideset{}{{}^{\prime}}{\sum}_{|J|=q}\alpha_{JK}(u_{\nu})_{J}\bigg)\ \tau_{K}\longrightarrow\sideset{}{{}^{\prime}}{\sum}_{|K|=q+1}f_{K}\ \tau_{K}\quad\text{in $\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q+1})$}. If we think pointwise, a consequence of ( 4.5 ) is that, for every x\in G fixed, \mathsf{T}(u_{\nu}(x))\longrightarrow f(x)\quad\text{in $\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})$}. By finite dimensionality of \mathbb{C}_{G}(G;\underline{\Lambda}^{q+1}) , we conclude that f(x)\in\ran\mathsf{T} for every x\in G . For each x\in G we define u(x)\doteq\mathsf{S}(f(x))\in\mathbb{C}_{G}(G;\underline{\Lambda}^{q}) : the assignment x\in G\longmapsto u(x)\in\mathbb{C}_{G}(G;\underline{\Lambda}^{q}) is obviously smooth – hence defines an element u\in\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) –, and satisfies \mathsf{T}(u(x))=f(x) for every x\in G . By the definition of \mathsf{S} , the coefficients of u with respect to the basis \tau_{J} with |J|=q are linear combinations with complex coefficients of the coefficients f , and since f\in\mathscr{K}(G;\underline{\Lambda}^{q+1}) the coefficients of u are also in \mathscr{K}(G) , and, therefore, u\in\mathscr{K}(G;\underline{\Lambda}^{q}) . We conclude from ( 4.4 ) that \mathrm{d}^{\prime}u=f . ∎ Remark 4.2 (A representation formula for H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G)) ) . Notice that, for each q , (4.6) \mathscr{K}(G;\underline{\Lambda}^{q})\cong\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q}) (this and other isomorphisms below are meant as topological linear). Indeed, since \mathbb{C}_{G}(G;\underline{\Lambda}^{q}) is a finite dimensional vector space, we may take \theta_{1}^{(q)},\ldots,\theta_{N_{q}}^{(q)} a linear basis for it and represent every element in \mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q}) uniquely as \sum_{j=1}^{N_{q}}f_{j}\otimes\theta_{j}^{(q)},\quad f_{j}\in\mathscr{K}(G), to which we assign the q -form \sum_{j=1}^{N_{q}}f_{j}\theta_{j}^{(q)}\in\mathscr{K}(G;\underline{\Lambda}^{q}) . This assignment is obviously linear, continuous, injective and, moreover, surjective: any u\in\mathscr{K}(G;\underline{\Lambda}^{q}) is written as ( 2.1 ) with u_{J}\in\mathscr{K}(G) each ordered multi-index J=(j_{1},\ldots,j_{q}) ; writing each \tau_{J}\in\mathbb{C}_{G}(G;\underline{\Lambda}^{q}) as a linear combination of \theta_{1}^{(q)},\ldots,\theta_{N_{q}}^{(q)} settles the claim. Continuity of its inverse is evident. Also, as far as ( 4.6 ) is concerned, \mathrm{d}^{\prime}:\mathscr{K}(G;\underline{\Lambda}^{q})\to\mathscr{K}(G;\underline{\Lambda}^{q+1}) is represented as 1\otimes\mathrm{d}^{\prime}:\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\longrightarrow\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1}), thanks to the obvious identity \mathrm{d}^{\prime}(f\theta)=f\mathrm{d}^{\prime}\theta whenever f\in\mathscr{K}(G) ; in that sense, ( 4.6 ) is a chain map, hence an isomorphism between differential complexes. Thus: \displaystyle H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G)) \displaystyle\cong\frac{\ker\{1\otimes\mathrm{d}^{\prime}:\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\longrightarrow\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})\}}{\ran\{1\otimes\mathrm{d}^{\prime}:\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q-1})\longrightarrow\mathscr{K}(G)\otimes\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\}} \displaystyle=\frac{\mathscr{K}(G)\otimes\ker\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\longrightarrow\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})\}}{\mathscr{K}(G)\otimes\ran\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q-1})\longrightarrow\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\}}. Finally, it follows from [ 4 , Proposition 2.18] that exactness of the short sequence of vector spaces \ran\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q-1})\to\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\}\xrightarrow{\text{incl.}}\\ \ker\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\to\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})\}\xrightarrow{\text{proj.}}H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G})\longrightarrow 0 entails exactness of \mathscr{K}(G)\otimes\ran\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q-1})\to\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\}\xrightarrow{1\otimes\text{incl.}}\\ \mathscr{K}(G)\otimes\ker\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\to\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})\}\xrightarrow{1\otimes\text{proj.}}\mathscr{K}(G)\otimes H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G})\longrightarrow 0, where, clearly, all the spaces involved are Fréchet and the arrows are continuous. The last arrow 1\otimes\text{proj.} in the sequence above is onto, and descends to the quotient as a continuous linear isomorphism \frac{\mathscr{K}(G)\otimes\ker\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\longrightarrow\mathbb{C}_{G}(G;\underline{\Lambda}^{q+1})\}}{\mathscr{K}(G)\otimes\ran\{\mathrm{d}^{\prime}:\mathbb{C}_{G}(G;\underline{\Lambda}^{q-1})\longrightarrow\mathbb{C}_{G}(G;\underline{\Lambda}^{q})\}}\cong\mathscr{K}(G)\otimes H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G}) whose inverse is continuous by the Open Mapping Theorem. Together, both isomorphisms prove that H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G))\cong\mathscr{K}(G)\otimes H^{q}_{\mathrm{d}^{\prime}}(\mathbb{C}_{G}). Next, we delve into the structure of the second factor in ( 4.2 ). Proposition 4.3 . Let q\in\{1,\ldots,n\} . If f\in\mathscr{R}(G;\underline{\Lambda}^{q}) is such that \mathrm{d}^{\prime}f=0 then there exists v\in\bar{\mathscr{R}}(G;\underline{\Lambda}^{q-1}) such that \mathrm{d}^{\prime}v=f . That is, the natural map (4.7) H^{q}_{\mathrm{d}^{\prime}}(\mathscr{R}(G))\longrightarrow H^{q}_{\mathrm{d}^{\prime}}(\bar{\mathscr{R}}(G)) is zero. Proof. Write a \mathrm{d}^{\prime} -closed f\in\mathscr{R}(G;\underline{\Lambda}^{q}) in terms of coefficients f_{J}\in\mathscr{R}(G) . For each J , we have f_{J}=\Delta_{\mathfrak{v}}g_{J} for some g_{J}\in\mathscr{C}^{\infty}(G) . Replacing g_{J} by \pi_{\bar{\mathscr{R}}}(g_{J}) , we may assume without loss of generality that g_{J}\in\bar{\mathscr{R}}(G) , and define g\doteq\sideset{}{{}^{\prime}}{\sum}_{|J|=q}g_{J}\ \tau_{J}\in\bar{\mathscr{R}}(G;\underline{\Lambda}^{q}), which obviously solves \Delta_{\mathfrak{v}}g=f . Moreover 0=\mathrm{d}^{\prime}f=\mathrm{d}^{\prime}(\Delta_{\mathfrak{v}}g)=\Delta_{\mathfrak{v}}(\mathrm{d}^{\prime}g), which implies that \mathrm{d}^{\prime}g\in\mathscr{K}(G;\underline{\Lambda}^{q+1}) . Since also \mathrm{d}^{\prime}g\in\bar{\mathscr{R}}(G;\underline{\Lambda}^{q+1}) , we conclude that \mathrm{d}^{\prime}g=0 , and therefore Lemma 2.3 ensures the existence of a v\in\bar{\mathscr{R}}(G;\underline{\Lambda}^{q-1}) solving \mathrm{d}^{\prime}v=\Delta_{\mathfrak{v}}g=f . ∎ 5. Globally solvable structures Theorem 5.1 . If \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G)\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1}) has closed range then H_{\mathrm{d}^{\prime}}^{q}(\bar{\mathscr{R}}(G))=0,\quad\forall q\in\{0,\ldots,n\}, and in that case the natural map (5.1) H^{q}_{\mathrm{d}^{\prime}}(\mathscr{K}(G))\longrightarrow H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G)) is a topological isomorphism; in particular, H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G)) is a Fréchet space. Moreover, every [f]\in H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G)) has a representative \pi_{\mathscr{K}}(f)\in\mathscr{K}(G;\underline{\Lambda}^{q}) . Proof. On the one hand, we know from Proposition 4.3 that the natural map ( 4.7 ) is always zero. On the other hand, the hypothesis of closedness of the range of \mathrm{d}^{\prime} is equivalent (as already mentioned [ 3 , Proposition 3.11] ) to the closedness of ( 3.1 ) in \mathscr{C}^{\infty}(G) , i.e., \mathscr{R}(G)=\bar{\mathscr{R}}(G) , in which case ( 4.7 ) is the identity: the first conclusion follows at once. Now, by the direct sum decomposition ( 4.2 ), our first conclusion entails that the map ( 5.1 ) is onto; we already knew that it was a continuous injection. Hence, ( 5.1 ) is a continuous linear isomorphism. Its inverse is precisely the map induced by \pi_{\mathscr{K}} in cohomology, hence continuous as well. The final claim is simply a restatement of the surjectivity of ( 5.1 ). ∎ Corollary 5.2 . If \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G)\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1}) has closed range then so does \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q-1})\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) for every q\in\{2,\ldots,n\} . Corollary 5.3 . If \mathcal{V} is globally hypoelliptic then every class in H^{q}_{\mathrm{d}^{\prime}}(\mathscr{C}^{\infty}(G)) has a representative in \mathbb{C}_{G}(G;\underline{\Lambda}^{q}) . Proof. The hypothesis is equivalent to global hypoellipticity of the first \mathrm{d}^{\prime} in the differential complex associated to \mathcal{V} ; that is, any f\in\mathscr{D}^{\prime}(G) satisfying \mathrm{d}^{\prime}f\in\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1}) is already smooth to start with. This implies [ 2 , Theorem 2.2] that the map \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G)\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1}) has closed range; a standard functional analytic argument further shows that its kernel \mathscr{K}(G) is of finite dimension. Since the latter is a subalgebra of \mathscr{C}^{\infty}(G) (by Leibniz rule), it must be equal to \mathbb{C}_{G} . The conclusion will follow from Theorem 5.1 . Indeed, let f\in\mathscr{K}(G) . Hence, f^{n}\in\mathscr{K}(G) for every n\in\mathbb{N} , and since \mathscr{K}(G) is finite dimensional there exists n_{0}\in\mathbb{N} such that 1,f,f^{2},\ldots,f^{n_{0}} are linearly dependent. This entails the existence of a polynomial P\in\mathbb{C}[z] such that P(f)=0 , meaning that f(G) consists of roots of P , which are finitely many. By connectedness, f(G) is a point, i.e., f is constant. ∎ It is a natural question to ask whether some kind of converse to Corollary 5.2 holds. We provide an affirmative answer assuming that \mathfrak{v} is commutative. Notice that this assumption does not preclude non-commutativity of G . Theorem 5.4 . Assume that \mathfrak{v} is Abelian. If \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q-1})\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) has closed range for some q\in\{2,\ldots,n\} , then \mathrm{d}^{\prime}:\mathscr{C}^{\infty}(G)\to\mathscr{C}^{\infty}(G;\underline{\Lambda}^{1}) has closed range. Proof. The key algebraic aspect of Abelian structures is the simplification of the expression ( 2.3 ) of \mathrm{d}^{\prime} thanks to the vanishing of the structural constants \alpha_{JK} . Indeed, by ( 4.3 ) we have, for an ordered multi-index K=(k_{0},\ldots,k_{q}) : \alpha_{JK}=(\mathrm{d}^{\prime}\tau_{J})(\mathrm{L}_{k_{0}},\ldots,\mathrm{L}_{k_{q}})=0 thanks to ( 2.4 ), the duality relations between \tau_{1},\ldots,\tau_{n} and \mathrm{L}_{1},\ldots,\mathrm{L}_{n} , and the fact that the latter is a family of pairwise commuting vector fields (our hypothesis). Therefore, if we write u\in\mathscr{C}^{\infty}(G;\underline{\Lambda}^{q}) in its canonical form ( 2.1 ), then (5.2) \mathrm{d}^{\prime}u=\sum_{j=1}^{n}\sideset{}{{}^{\prime}}{\sum}_{|J|=q}(\mathrm{L}_{j}u_{J})\ \tau_{j}\wedge\tau_{J}. The proof of the theorem starts by assuming, otherwise, that the range of \Delta_{\mathfrak{v}}:\mathscr{C}^{\infty}(G)\to\mathscr{C}^{\infty}(G) is not closed. By results in [ 1 ] , given C,\rho>0 there exist \lambda\in\sigma(\Delta_{G}) and \phi\in E_{\lambda}\cap\left(\ker\Delta_{\mathfrak{v}}\right)^{\perp} such that \|\phi\|_{L^{2}(G)}=1 and (5.3) \Big(\sum_{j=1}^{n}\|\mathrm{L}_{j}\phi\|_{L^{2}(G)}^{2}\Big)^{\frac{1}{2}}0. Picking 0\lambda_{1} : otherwise, \phi_{2}\in\mathcal{S}_{\lambda_{1}} , hence C_{2}(1+\lambda_{2})^{-2}