Distortion of mappings and \(L_{q,p}\)-cohomology (Q846892)

The \(L_{q,p}\)-cohomology of degree \(k\) of a Riemannian manifold \(M\), denoted by \(H_{q,p}^k(M)\), is the quotient of the space of \(p\)-integrable closed differential \(k\)-forms on \(M\) over the subspace of exact forms with a \(q\)-integrable primitive; here, \(p\)-integrability of a differential form \(\omega\) means that \(|\omega|\in L^p(M)\). It has appeared as a natural generalization of \(L_p\)-cohomology, which in turn is a natural generalization of \(L_2\)-cohomology. \(L_{q,p}\)-cohomology has connections with Sobolev type inequalities and conformal geometry. The paper exhibits a class of diffeomorphisms that induce homomorphisms in \(L_{q,p}\)-cohomology, which can be understood as the kind of functoriality satisfied by this cohomology. Given a diffeomorphism between Riemannian manifolds \(f:M\to\widetilde{M}\) and \(0\leq k\leq n\), let \(\sigma_k(f)\) be the function on \(M\) defined by \(\sigma_k(f,x)=\sum_{i_1<\dots<i_k}\lambda_{i_1}(x)\cdots\lambda_{i_k}(x)\), where \(\lambda_1(x),\dots,\lambda_n(x)\) are the eigenvalues of \(\sqrt{(df_x)^*(df_x)}\). Let \(J_f\) be the function on \(M\) defined by the Jacobian of \(f\) at each point. It is said that \(f\) has bounded \((s,t)\)-distortion in degree \(k\) if \(\sigma_k(f)\,J_f^{-1}\in L^t(M)\), and the family of these functions is denoted by \(BD_{s,t}^k(M,\widetilde{M})\). In particular, \(BD_{n,\infty}^1(M,\widetilde{M})\) (\(n=\dim M\)) consists of the quasi-conformal diffeomorphisms, and \(BD_{1,\infty}^n(M,\widetilde{M})\) consists of all diffeomorphisms. It is proved in the paper that any diffeomorphism \(f\in BD_{\tilde q,u}^{n-k+1}(M,\widetilde{M})\cap BD_{\tilde p,t}^k(M,\widetilde{M})\) induces a homomorphism \(f^*:H_{\tilde q,\tilde p}^k(\widetilde{M})\to H_{q,p}^k(M)\) if \(p\leq\tilde p<\infty\), \(q\leq\tilde q\), \(t=\frac{p}{\tilde p-p}\) and \(u=\frac{q}{\tilde q-q}\). Moreover it is shown that, if there is a diffeomorphism \(f\in BD_{\tilde q',r}^{n-k+1}(M,\widetilde{M})\cap BD_{\tilde p,t}^k(M,\widetilde{M})\) with \(p\leq\tilde p<\infty\), \(q\leq\tilde q>1\), \(\tilde q'=\frac{\tilde q}{\tilde q-1}\) and \(r=\frac{q(\tilde q-1)}{q-\tilde q}\), then \([f^*\omega]=0\) in \(H_{q,p}^k(M)\) implies \([\omega]=0\) in \(H_{\tilde q,\tilde p}^k(\widetilde{M})=0\); in particular, \(H_{q,p}^k(M)=0\) implies \(H_{\tilde q,\tilde p}^k(\widetilde{M})=0\). This second result is used to prove the following vanishing result for a manifold with a cusp. Take \(M=\mathbb{R}^n\) with its standard Euclidean metric \(g\), and take \(\widetilde{M}=\mathbb{R}^n\) with a metric \(\tilde g\) such that, in polar coordinates, \(\tilde g=dr^2+e^{-2r}h\) for large enough \(r\), where \(h\) is the standard metric on \(S^{n-1}\). It is shown that the identity map \(f:M\to\widetilde{M}\) belongs to \(BD_{s,t}^m(M,\widetilde{M})\) if \(s>\frac{n-1}{m-1}\) and \(0<t\leq\infty\), obtaining that \(H_{\tilde q,\tilde p}^k(\widetilde{M})=0\) if \(\tilde q<\frac{n-1}{k-1}<\tilde p\).