\documentclass[11pt]{article}\usepackage{amssymb,latexsym,amsmath,amsfonts,amsthm}\newtheorem{thm}{Theorem}[section]\newtheorem{lem}[thm]{Lemma}\newtheorem{prop}[thm]{Proposition}\newtheorem{cor}[thm]{Corollary}\newtheorem{defn}{Definition}[section]\theoremstyle{remark}\newtheorem{rem}{\textsc{Remark}}[section]% MATH -----------------------------------------------------------\newcommand{\abs}[1]{\left\vert#1\right\vert}\newcommand{\Om}{\Omega}\newcommand{\RR}{I\!\!R}\newcommand{\dsp}{\displaystyle}\def\m{\mu}\def\Lam{\Lambda}\def\a{\alpha}\def\b{\beta}\def\n{\nu}\def\Om{\Omega}\def\wto{\rightharpoonup}\def\r{\rho}\def\dint{-\kern-1.1em\int}\def\om{\omega}\def\aplim{\mathop{\rm aplim}}\def\lam{\lambda}\def\newline{\hfill\break}\def\e{\varepsilon}\def\d{\delta}\def\g{\gamma}\def\lam{\lambda}\def\t{\tau}\def\s{\sigma}\catcode`@=11\newbox\tr@tto\setbox\tr@tto=\hbox{{\count0=0\dimen0=-,9pt\dimen1=1,1pt\loop\ifnum    \count0<11 \advance \count0 by1 \vrule width.51pt height\dimen1    depth\dimen0\kern-0.17pt\advance\dimen0 by-0.05pt\advance\dimen1    by0.1pt\repeat \loop\ifnum\count0<21\advance \count0 by1 \vrule    width.6pt height\dimen1 depth\dimen0\kern-0.2pt \advance\dimen0    by-0.1pt\advance\dimen1 by 0.05pt\repeat}}\def\medint{\displaystyle\copy\tr@tto\kern-10.4pt\int} \catcode`@=12\def\bull{\vrule height .9ex width .8ex depth -.1ex }%\newcommand\baselineskip=16pt\begin{document}\begin{center}{\bf ISOPERIMETRIC TYPE INEQUALITIES FOR\\DIFFERENTIAL FORMS ON MANIFOLDS }\end{center}\medskip\begin{center}{ Flavia Giannetti * Antonia Passarelli di Napoli *}\end{center}\medskip\begin{center}{\small  *Dipartimento di Matematica e Applicazioni "R. Caccioppoli"}{\small Universit\`{a} di Napoli ``Federico II", via Cintia - 80126 Napoli}{\footnotesize e-mail:giannett@unina.it\qquad antonia.passarelli@unina.it}\end{center}\vskip2cm\bigskip\bigskip\noindent {\footnotesize  ABSTRACT. Let $X$ be a smooth oriented Riemannian $n$-manifold without   boundary and $(\Phi,\Psi)\in {\cal L}^p(\wedge ^l X)\times{\cal L}^r(\wedge ^{n-l} X)$ ,$\frac {1}{p}+\frac {1}{r}=1+\frac {1}{n}$ , be a  pair of closed differential forms. We prove an isoperimetric type inequality for such differential forms under suitable assumptions. As an application we derive H\"older continuity for solutions of Hodge systems.}\bigskip\noindent {\footnotesize {\bf AMS Classifications.}  58A10; 58A14 ; 58J99.}\bigskip\noindent {\footnotesize {\bf Key words.} {\it Wedge Products; Isoperimetric Inequality;H\"older continuity.}}\bigskip\bigskip\bigskip\vskip2cm\section{Introduction}The familiar geometric form of the isoperimetric inequality reads as$$n^{n-1}\omega_{n-1}|U|^{n-1}\leqslant |\partial U|^{n}$$where $|U|$ stands for the volume of a region $U\subset \RR^{n}$, while $|\partial U|$ is its $(n-1)$-dimensional surface area.  Now, if $f:R\to U$ is a smooth diffeomorphism of a ``regular'' domain $Q\subset \RR^{n}$onto $U$ then $|U|=\int_{Q}{\cal J}(x,f)dx$, where ${\cal J}(x,f)$ is the Jacobian determinant of $Df$, while $|\partial U|$ is dominated by $\int_{\partial Q}|Df|^{n-1}dx$ or $\int_{\partial Q}|D^{\sharp}f|dx$. Thus we obtain the following integral form of the isoperimetric inequalities\begin{equation}\label{isopuno}    \Bigg|\int_{Q}{\cal J}(x,f)dx\Bigg|\leqslant C(n)\Bigg(\int_{\partial Q}|Df|^{n-1}dx\Bigg)^{\frac{n}{n-1}}\end{equation}and\begin{equation}\label{isopdue}    \Bigg|\int_{Q}{\cal J}(x,f)dx\Bigg|\leqslant C(n)\Bigg(\int_{\partial Q}|D^{\sharp}\!f|dx\Bigg)^{\frac{n}{n-1}},\end{equation}valid also for all smooth mappings $f:Q\to \RR^{n}$, not necessarily diffeomorphisms.Reshetnyak  $\cite{[13]}$ established sharp H\"older continuity for  mappings of bounded distortion  extending  ideas of Morrey \cite{[7]}. This required  the isoperimetric inequality (\ref{isopdue}) for mappings in the Sobolev class $W^{1,n}$. This inequality has later been explored  in  nonlinear elasticity \cite{[8]},\cite{[9]}.Let us note that given a mapping $f:\RR^{n}\to\RR^{n}$ in $W^{1,n}$, we have two vector fields associated with $f$. The first one is curl free while the second is divergence free. These are, respectively$$E=\nabla f^{1}\qquad{\rm and}\qquad  B=(B^{1},\dots, B^{n})$$ where$$B^{j}=(-1)^{j+1}\frac{\partial (f^{2},\dots,f^{n})}{\partial (x^{1},\dots,x^{j-1},x^{j+1},\dots,x^{n})}$$Their scalar product $\langle B, E\rangle$, which equals exactly the Jacobian determinant of $Df$, is integrable for $f\in W^{1,n}(\RR^{n})$.Now, any pair $(B,E)$ of vector fields with div $B$=0 and curl $E$=0 will be  referred to as {\it div-curl couple}. By analogy to the previous  discussion the scalar product $J=\langle B,  E\rangle$ will be called the Jacobian of $(B,E)$.  Recently  \cite{[1]} an isoperimetric inequality of  type (\ref{isopuno}) for div-curl couples $(B,E)\in L^{p}\times  L^{q}(\RR^{n}\times\RR^{n})$ has been proved, under the  assumption that $p,q$ are H\"older conjugate exponents. This makes  their Jacobian integrable.  As a  consequence the authors derived a H\"older  continuity property  of $p$-harmonic functions.   Actually,  local integrability of the Jacobian of   an orientation preserving  mapping $f$ was established  under  somewhat weaker assumption, i.e. $f\in W^{1,\frac {n^{2}}{n+1}}(\RR^{n})$ and $$\liminf_{t\to\infty}t^{{n+1\over n}}\int_{|Df|>t}|Df(x)|^{\frac  {n^{2}}{n+1}}dx=0\,.$$ Later  \cite{[11]}    an isoperimetric inequality of the type (\ref{isopdue})  for such  mappings $f$  has also been proved.  In this spirit one might try to obtain an  isoperimetric type inequality  for  a  wedge product of closed differential forms on a smooth Riemannian manifold  analogous to that in \cite{[1]}. However, attempts to weaken the assumptions on the  integrability of the forms fail when the wedge product  doesn't obey  the rule of integration by parts. Actually,  this  problem translates into the  question how the distributional  wedge  product  relates to the pointwise product.  In this paper we were able to identify these two  quantities. Then we   proved  an isoperimetric type inequality  under assumptions similar to   those in \cite{[3]}.  Let $X$ be a smooth oriented Riemannian $n$-manifold without   boundary. A  pair of closed differential   forms  $\Phi\in {\cal L}^p(\wedge ^l X)\cap$ ker $d$ and $\Psi\in{\cal L}^r(\wedge ^{n-l} X)\cap$ ker $d$, where  $1< p,r<\infty$ satisfySobolev's relation $\frac {1}{p}+\frac {1}{r}=1+\frac {1}{n}$,  will be called   an {\it admissible pair}  if $\Phi\wedge \Psi \ge 0$\footnote{Throughout this text we fix an orientation of $X$, that is a nonvanishing $n$-form $dx\in {\cal C}^{\infty}(\wedge ^n X)$. We then say that an $n$-form $\Theta$ on $X$ is nonnegative if $\Theta=\lambda dx$ for some nonnegative function $\lambda$.} and\begin{equation}\label{acca}\liminf_{t\to\infty}t^{{1\over n}}\int_{H>t}H(x)dx=0\end{equation}where $H=|\Phi|^{p}+|\Psi|^{r}$.    The reader may wish to notice that this equation is automatic if $H\in {\cal L}^{n+1\over n}( X)$. However, there are many interesting cases in which (\ref{acca}) holds for $H\not \in {\cal L}^{n+1\over n}( X)$ Our main result is contained in the following  \medskip\begin{thm}\label{dis} Let $(\Phi, \Psi)$ be an admissible pair.   Given $x\in X$, for almost every ball ${\bf B}={\bf B}(x,\rho)$ in $X$    we have   \begin{equation}\label{tesi}\int_{\bf B}\Phi\wedge \Psi<\!\!< \Big(\int_{\partial{\bf B}}|\Phi|^{p}d{\cal H}^{n-1}\Big)^{\frac{1}{p}} \Big(\int_{\partial{\bf B}}|\Psi|^{s}d{\cal H}^{n-1}\Big)^{\frac{1}{s}}\end{equation}provided $ 1< s=\frac{p(n-1)}{np-n+1}$ and $0<\rho\le R_{X}$, where $R_{X}$ depends only on $X$.\end{thm}The symbol $<\!\!<$ means that inequality (\ref{tesi}) holds with a constant depending on $X$.\noindent As a first consequence we obtain an isoperimetric inequality of the type (\ref{isopuno})  when the target manifold has anontrivial cohomology ${\cal H}^{l}( Y)$, for some $1< l<n-1$. Secondly, as in \cite{[1]}, we derive a  H\"older  continuity property for solutions of Hodge systems.\medskip\section{Preliminaries}{\bf 2.1 Differential forms}\bigskipOur ambient space is an oriented compact (without boundary) smooth Riemannian manifold $X$ of dimension $n> 2$.\noindent Fix an atlas $\cal A$ consisting of a finite number of coordinate charts $(\Omega_{i}, \vartheta_{i})$, where $\vartheta_{i}:\Omega_{i} \to \RR^{n}$ is a $\cal C^{\infty}$- diffeomorphism of the open region $\Omega_{i}\subset X$ onto $\RR^{n}$ and choose a ${\cal C}^{\infty}$- partition of unity $\{\chi_{i}\in {\cal C}_{0}^{\infty}(\Omega_{i}):\,\, \sum \chi_{i}=1\}$ subordinate to $\cal A$.\noindent Introduce a radius $R=R_X$ such that all concentric balls ${\bf B}(x,\rho)\subset {\bf B}(x,3\rho)\subset X$, with $\rho\le R$ lay in the same coordinate region $\Omega_{i}$ of the fixed atlas  $\cal A$. Such balls share many properties of the Euclidean balls.Let us denote by ${\cal C}^{\infty}(\wedge ^l X)$, $ 1\le l \le n=dim X$, the space of smooth $l$-forms on $X$. Consider the exterior derivative $$d:{\cal C}^{\infty}(\wedge ^l X)\to {\cal C}^{\infty}(\wedge ^{l+1} X)$$and its formal adjoint $$d^{*}=(-1)^{nl+1}*d*:{\cal C}^{\infty}(\wedge ^{l+1} X)\to {\cal C}^{\infty}(\wedge ^lX)$$where $*:{\cal C}^{\infty}(\wedge ^l X)\to {\cal C}^{\infty}(\wedge ^{n-l} X)$ denotes the Hodge star duality operator. The duality between $d$ and $d^{*}$ is emphasized in the integration by parts $$ \int_{X}\langle d\varphi, \psi\rangle =\int_{X}\langle \varphi, d^{*}\psi\rangle$$for $\varphi\in {\cal C}^{\infty}(\wedge ^l X)$ and $\psi\in {\cal C}^{\infty}(\wedge ^{l+1} X)$.We denote  by ${\cal L}^p(\wedge ^l X)$ the space of all measurables $l$-forms $\omega$ for which $$||\omega||_{p}=\Big(\int_{X}|\omega|^{p}\Big)^{1\over p}<+\infty \qquad {\rm for}\,\,\, 1\le p<+\infty$$and$$||\omega||_{\infty}={\rm ess}\sup |\omega|\qquad {\rm for}\,\,\,  p=\infty$$In order to define the Sobolev space ${\cal W}^{1,p}(\wedge ^l X)$, $1\le p\le +\infty$, we decompose an $l$-form $\omega$ as $\omega=\sum \omega_{i}$ where $\omega_{i}=\chi_{i}\omega$ and consider the pullback of $\omega_{i}$ via the mapping $\vartheta_{i}^{-1}:\RR^{n}\to \Omega_{i}$$$\omega_{i}^{\sharp}(x)=\sum_{I}\omega_{i}^{I}(x)dx^{I}\qquad x\in \theta_{i}(\Omega_{i})\subset \RR^{n} $$The sum is taken over all ordered $l$-tuples $I=(i_{1},\ldots, i_{l})$ and $dx^{I}=dx_{i_{1}}\wedge\ldots\wedge  dx_{i_{l}}$. If each function $\omega_{i}^{I}$ has generalized partial derivative $\partial\omega_{i}^{I}\in {\cal L}^p$ then the Sobolev norm of $\omega$ is defined by$$||\omega||_{1,p}=\sum_{i} ||\omega_{i}||_{1,p}$$A differential form $\varphi \in {\cal L}^p(\wedge ^l X)$ is said to be closed (in the sense of distributions), in writing $d\varphi=0$, if $$ \int_{X}\langle \varphi, d^{*}\psi\rangle =0$$ for every test form $\psi\in {\cal C}^{\infty}(\wedge ^{l+1} X)$.Similarly we define the coclosed forms, $ d^{*}\psi=0$. Finallya form $h\in {\cal L}^p(\wedge ^l X)$ which is both closed and coclosed is called  a harmonic field of degree $l$. Note that the space of all harmonic fields of degree $l$ is finite dimensional and consists of ${\cal C}^{\infty}$-smooth forms.We shall use the following Hodge decomposition of a differential form $\omega \in {\cal L}^p(\wedge ^l X)$  (see \cite{[6]}):\begin{thm}\label{hodge}    For $1<p<\infty$, and $1\le l\le n$ every $\omega \in {\cal L}^p(\wedge ^l X)$ can be decomposed as$$\omega = d\alpha+d^{*}\beta +h$$where $\alpha,\beta$ are in the Sobolev classes ${\cal W}^{1,p}(\wedge ^{l-1} X)$ , ${\cal W}^{1,p}(\wedge ^{l+1} X)$ respectively and $h $ is a harmonic field of degree $l$.\end{thm}\bigskip\noindent {\bf 2.2 Weakly differentiable mappings}\bigskipOur applications will be concerned with weakly differentiable mappings $f:X\to Y$ where $X$ and $Y$ are  oriented smooth Riemannian manifolds  of dimension $n\ge 2$ without boundary. \noindent In order to define the Sobolev class ${\cal W}^{1,p}(X,Y)$ we appeal to the well known Theorem of J. Nash (\cite{[10]}) which ensures that $Y$ can be ${\cal C}^{\infty}$-isometrically imbedded in the euclidean space $\RR^{n}$.\noindent Namely, we say that $f: X\to \RR^{n}$ lies in ${\cal W}^{1,p}(X,Y)$ if each coordinate function $f^{i}:X\to \RR$ belongs to the  usual Sobolev space ${\cal W}^{1,p}(X)$ and $f(x)\in Y$ for almost every $x\in X$.To every ${\cal C}^{\infty}$-smooth $l$-form $\omega$ in the target  space there corresponds its pullback $f^{\sharp}\!\omega$ via $f$. The pointwise estimate $$ |f^{\sharp}\!\omega|<\!\!< |Df|^{l}$$relates the degree of integrability of $f^{\sharp}\!\omega$ to the one of $Df$.\noindent Here and in what follows the symbol $<\!\!<$ means that the inequality holds with a constant depending  on the manifold $X$.\noindent Note that the commutation rule $$ f^{\sharp}\circ d=d\circ f^{\sharp}$$holds provided $f:X \to Y$ belongs to $C^{\infty}$.Given a function $h\in {\cal L}^p(X)$  define the maximal function of $h$ as  $${\bf M}_{p}h(x)=\sup\Big\{\Big(\medint_{\bf B}|h(x)|^{p}dx\Big)^{1\over p};\,\, x\in {\bf B}\subset X\Big\}$$where the supremun runs over all balls ${\bf B}\subset X$ containing $x$.\noindent As in the Euclidean setting, the following weak type estimate holds (see \cite{[4]})\medskip\begin{prop}\label{propmax}    For every $h\in {\cal L}^p(X)$  we have\begin{equation}\label{massimale}\int_{{\bf M}_{p}h>2t}dx<\!\!< {1\over t^{p}}\int_{|h|>t}|h(x)|^{p}dx\end{equation}for all $t>0$.\end{prop}\noindent As a consequence of the weak type estimate at (\ref{massimale}) we obtain\medskip\begin{cor}\label{cormax} Let $\{h_{j}\}$ converges to $h$ in ${\cal L}^p(X)$, then\begin{equation}    \lim_{j \to \infty} \int_{M_{p}h_{j}>\lambda} dx= \int_{M_{p}h>\lambda} dx\end{equation}for every $\lambda>0$. Moreover $\{M_{p}h_{j}\}$ contains a subsequence converging to $M_{p}h$ almost everywhere.\end{cor}\bigskip\section{Weak wedge products}\bigskipOur intention here is to identify the distributional wedge product of closed forms with their pointwise product. As a matter of fact, a significant part of this section comes from \cite{[4]}.\noindent Consider closed differential forms $\Phi\in {\cal L}^p(\wedge ^l X)\cap$ ker $d$ and $\Psi\in{\cal L}^r(\wedge ^k X)\cap$ ker $d$, where $1\le k,l<n,\, k+l=n$ and $1< p,r<\infty$. Firstassume that $p$ is H\"older conjugate to $r$. Thus $\Phi\wedge \Psi$ is integrable and, therefore,defines a Schwartz distribution of order zero\begin{equation}(\Phi\wedge \Psi)[\eta]=\int_X \eta (\Phi\wedge \Psi)\qquad\qquad \mbox{for}\,\, \eta \,\in {\calC}^\infty (X) \end{equation}But we can do much better if we use the following Hodge decomposition, see Theorem \ref{hodge}, \begin{equation}\Phi= d\varphi+\theta\end{equation}where the exact component $d\varphi\in {\cal L}^p(\wedge ^l X)$ and the harmonic field $\theta\in {\cal C}^\infty(\wedge ^l X)$.Then, using integration by parts, we obtain  the following formula \begin{eqnarray}(\Phi\wedge \Psi)[\eta]& = &\int_X \eta (d\varphi\wedge \Psi)+\int_X \eta (\theta\wedge \Psi)\cr\cr&  =& \int_X \eta (\theta\wedge \Psi)- \int_X d\eta \wedge(\varphi\wedge \Psi)\end{eqnarray}This latter integral converges whenever $ \varphi\wedge \Psi$ is integrable. Assume now that $1< p,r<\infty$ are Sobolev conjugate exponents, that is $$ \frac{1}{p}+\frac{1}{r}=1+\frac{1}{n}$$This condition forces that one of the exponents is less than $n$, for instance $1< p<n$. Then, by Sobolev imbedding Theorem, we find that $\varphi\in {\cal L}^{\frac{np}{n-p}}(\wedge ^{l-1} X)$, where the exponent $\frac{np}{n-p}$ is exactly H\"older conjugate to $r$  so that $\varphi\wedge \Psi \in {\cal L}^1(\wedge ^{n-1} X)$.In this way we are justified  to give the  definition of the {\it distributional wedge product}.\begin{defn}  Let $\Phi\in {\cal L}^p(\wedge ^l X)\cap$ ker $d$ and $\Psi\in{\cal L}^r(\wedge ^{n-l} X)\cap$ ker $d$, where  $1< p,r<\infty$ areSobolev conjugate exponents. The distribution $\Phi\wedge \Psi$ operates on a test function $\eta\in {\cal C}^\infty( X)$ by the rule\begin{equation}\label{limit}    (\Phi\wedge \Psi)[\eta]=\lim_{j\to \infty}\int_X \eta (\Phi_{j}\wedge \Psi_{j})\end{equation}where $\Phi_{j}\in {\cal C}^\infty(\wedge ^l X)$ and $\Psi_{j}\in {\cal C}^\infty(\wedge  ^{n-l} X)$ are closed forms converging to $\Phi$ and $\Psi$ in ${\cal L}^p(\wedge ^l X)$ and ${\cal L}^r(\wedge  ^{n-l} X)$, respectively.\end{defn}\medskip\bigskipWe shall use  the following two Lemmas.\medskip\begin{lem}\label{lemma1}Let $\Phi\in {\cal C}^\infty(\wedge ^l X)$  and $\Psi\in{\cal C}^\infty(\wedge  ^{n-l} X)$ be closed differential forms and let  $1< p,r<\infty$ beSobolev conjugate exponents. Then for every nonempty open set $\Omega\subset X$ and every test function $\eta\in {\cal C}^\infty(X)$ we have\begin{equation}    \Bigg|\medint_{\Omega}\eta (\Phi\wedge \Psi)\Bigg|<\!\!<\Vert\eta\Vert_{{\cal C}^\infty(\Omega)}\Vert{\bf M}_{p}\Phi\Vert_{{\cal L}^{\infty}(X\setminus\Omega)}\Vert{\bf M}_{r}\Psi\Vert_{{\cal L}^{\infty}(X\setminus\Omega)}\end{equation}\end{lem}\bigskip\begin{lem}\label{lemma2}Let $\Phi\in {\cal C}^\infty(\wedge ^l X)$  and $\Psi\in{\cal C}^\infty(\wedge  ^{n-l} X)$ be closed differential forms, let  $1< p,r<\infty$ beSobolev conjugate exponents and let $\lambda >0$. Then, setting $H=|\Phi|^{p}+|\Psi|^{r}$, we have\begin{equation}\label{distr}    \Bigg|(\Phi\wedge \Psi)[\eta]-\int_{X\setminus\Omega}\eta (\Phi\wedge     \Psi)\Bigg|<\!\!<\lambda^{1+{1\over n}}|\Omega|\,\Vert\eta\Vert_{{\cal C}^\infty(X)}\end{equation}where $\Omega=\{x\in X:\,\,  {\bf M}H(x)>\lambda\}$.\end{lem}\medskip\medskip\noindent {\it Proof:}\quad The left hand side of the inequality (\ref{distr}) reduces to \begin{equation}\Bigg|\int_{\Omega}\eta(\Phi\wedge \Psi)\Bigg|\end{equation}Using Lemma \ref{lemma1} and the definition of $\Omega$ we get\begin{eqnarray}\Bigg|\int_{\Omega}\eta(\Phi\wedge \Psi)\Bigg|&<\!\!<&|\Omega|\,\Vert\eta\Vert_{{\cal C}^\infty(X)}\Vert{\bf M}_{p}\Phi\Vert_{{\cal L}^{\infty}(X\setminus\Omega)}\Vert{\bf M}_{r}\Psi\Vert_{{\cal L}^{\infty}(X\setminus\Omega)}\cr\cr&<\!\!<&|\Omega|\,\Vert\eta\Vert_{{\cal C}^\infty(X)}\Vert{\bf M}H\Vert^{1\over p}_{{\cal L}^{\infty}(X\setminus\Omega)}\Vert{\bf M}H\Vert^{1\over r}_{{\cal L}^{\infty}(X\setminus\Omega)}\cr\cr&<\!\!<&|\Omega|\,\Vert\eta\Vert_{{\cal C}^\infty(X)}\lambda^{1\over p}\lambda^{1\over r}=|\Omega|\,\Vert\eta\Vert_{{\cal C}^\infty(X)}\lambda^{1+{1\over n}}\end{eqnarray}Now, we are in position to prove \medskip\begin{thm}Let $(\Phi,\Psi)$  be an admissible pair of differential forms. Then there are measurable sets $X_{1}\subset X_{2}\ldots \subset X$ whose union is $X$ and $\Phi\wedge \Psi$ is integrable over each $X_{j}$, such that \begin{equation}(\Phi\wedge \Psi)[\eta]=\lim_{j\to \infty}\int_{X_{j}}\eta(\Phi\wedge \Psi)\end{equation}for every $\eta\in {\cal C}^\infty(X)$.\end{thm}\medskip\medskip\noindent {\it Proof:}\quad Observe that $\Phi$ and $\Psi$ satisfy inequality (\ref{distr}) for all but a countable set of the parameters $\lambda>0$. Indeed, there exist sequences of closed forms $\Phi_{j}\in {\cal C}^\infty(\wedge ^l X)$  and $\Psi_{j}\in{\cal C}^\infty(\wedge  ^{n-l} X)$ converging to $\Phi$ and $\Psi$ in ${\cal L}^p(\wedge ^l X)$ and ${\cal L}^r(\wedge  ^{n-l} X)$, respectively. We need only to justify  a passage to the limit in the inequalities\begin{equation}    \Bigg|(\Phi_{j}\wedge \Psi_{j})[\eta]-\int_{{\bf M}H_{j}\le \lambda}\eta     (\Phi_{j}\wedge     \Psi_{j})\Bigg|<\!\!<\lambda^{1+{1\over n}}||\eta||_{{\cal C}^\infty(X)}\int_{{\bf M}H_{j}> \lambda}dx \end{equation}Letting $j\to\infty$ the integrals in the right hand side converge to $\int_{{\bf M}H> \lambda}dx $, by Corollary \ref{cormax}. Of course, $(\Phi_{j}\wedge \Psi_{j})[\eta]$ converges to $(\Phi\wedge \Psi)[\eta]$ by the definition of the distributional wedge product. Moreover, since ${\bf M}H_{j}$ converges to ${\bf M}H$ a.e. in X, the characteristic functions $\Xi_{j}$ of the sets $\{x\in X:\,\, {\bf M}H_{j}\le \lambda\} $ converge almost everywhere to the characteristic function $\Xi$ of the set $\{x\in X:\,\, {\bf M}H\le \lambda\} $. Therefore if we write the integral in the left hand side as$$ \int_{X}\eta     (\Phi_{j}\wedge     \Psi_{j})\Xi_{j}$$and observe that the integrands are uniformly bounded by $\lambda^{1+{1\over n}}||\eta||_{{\cal C}^\infty(X)}$, we can pass to the limit thank to the Dominated Convergence Theorem.   Let us consider a sequence $t_{j}$ increasing to infinity such that \begin{equation}\label{assu}\liminf_{j\to\infty}t_{j}^{{1\over n}}\int_{H>t_{j}}H(x)dx=0\end{equation}and define the sets $X_{j}$ as$$X_{j}=\{x\in X:\,\, {\bf M}H\le 2t_{j}\} $$We make use of the estimate in Lemma \ref{lemma2} with $\lambda=2t_{j}$\begin{eqnarray}\Bigg|(\Phi\wedge \Psi)[\eta]-\int_{X_{j}}\eta     (\Phi\wedge     \Psi)\Bigg|&<\!\!<&\Vert\eta\Vert_{{\cal C}^\infty(X)} t_{j}^{1+{1\over n}}\int_{{\bf M}H> 2t_{j}}dx\cr\cr&<\!\!<&\Vert\eta\Vert_{{\cal C}^\infty(X)} t_{j}^{{1\over n}}\int_{H> t_{j}}H(x)dx\end{eqnarray}The latter inequality follows by Proposition \ref{propmax}.Letting $t_{j}$ go to $\infty$, we conclude using  (\ref{assu}).\bigskipUsing a monotone convergence argument, from previous Theorem we obtain\medskip\begin{thm}Let $(\Phi,\Psi)$  be an admissible pair of differential forms. Then \begin{equation}(\Phi\wedge \Psi)[\eta]=\int_{X}\eta(\Phi\wedge \Psi)\end{equation}for every $\eta\in {\cal C}^\infty(X)$.\end{thm}\medskipFix a nonnegative $\eta\in {\cal C}^\infty(\RR^{n})$ supported in the closed unit ball  and having integral 1. We shall denote by   $\eta_{t}$, $t>0$, the usual sequence of mollifiers, i.e.  $\eta_{t}(x)=t^{-n}\eta(t^{-1}x)$. The convolution between $\Phi\wedge \Psi $ and $\eta_{t}$ is defined by the rule\begin{equation}\label{conv}(\Phi\wedge \Psi)\star\eta_{t}(x)=\sum_{\Omega_{i}\in{\cal A}}\chi_{i}(x)\int_{\Omega_{i}}\eta_{t}(\vartheta_{i}(x)-\vartheta_{i}(z))J(z,\vartheta_{i}(x))\Phi\wedge \Psi(z)\end{equation}where $\{\chi_{i}, i\in I\}$ is the  partition of unity subordinate to the fixed atlas ${\cal A}$ consisting of the coordinate charts $(\Omega_{i},\vartheta_{i})$.Note that $(\Phi\wedge \Psi)\star\eta_{t}$ converges to $\Phi\wedge \Psi$ in the sense of distributions, as $t\to 0$.Then if $(\Phi,\Psi)$  is an admissible pair of differential forms we have\begin{equation}\label{conv}(\Phi\wedge \Psi)(x)=\lim_{t\to 0}(\Phi\wedge \Psi)\star\eta_{t}(x)\end{equation}A simple limiting argument implies  the following rule of integration by parts for admissible pairs of differential forms \begin{equation}\label{parts}\int_X \eta (\Phi\wedge \Psi)=\int_X d\eta \wedge(\varphi\wedge \Psi)+\int_X \eta (h\wedge \Psi)\end{equation}where $\Phi=d\varphi+ h$.\bigskip\bigskip\section{Isoperimetric type inequalities}\bigskip\bigskipIn the Euclidean case it is possible to derive  isoperimetric type inequalities  from estimates of the  div-curl  product $\langle  B,E\rangle$ under suitable assumptions on the degree ofintegrability of $B$ and $E$ (see \cite{[1]}). In our context this idea translates in Theorem \ref{dis}, whose proof is the following\medskip\noindent {\it Proof:}\quad Since ${\bf B}$  has trivial cohomology, then every closed form $\Phi$ is also exact on ${\bf B}$, i.e. there exists a $(l-1)$-form $\varphi$ such that $d\varphi=\Phi$.\noindent Moreover, from the assumption on $s$ weinfer that $1< p<n-1$. Then there exists a closed $(l-1)$-form $\varphi_{0}$ such that $\varphi-\varphi_{0}\in {\cal W}^{1,p^{*}}(\Lambda^{l-1}X)$ and the following Sobolev-Poincar\'e inequality holds\begin{equation}\Big( \int_{\partial {\bf B}}|\varphi-\varphi_{o}|^{p^{*}}d{\cal H}^{n-1}\Big)^{1\over p^{*}}<\!\!<\Big( \int_{\partial {\bf B}}|d\varphi|^{p}d{\cal H}^{n-1}\Big)^{1\over p}\end{equation} where $ p^{*}=\frac{(n-1)p}{n-1-p}$ and ${\cal H}^{n-1}$ denotes the $(n-1)$-dimensional  measure on $\partial {\bf B}$.By the rule of integration by parts at (\ref{parts})  and the assumption $\Psi\in$ ker $d$ we have\begin{eqnarray}\int_{\bf B}\Phi\wedge \Psi&=& \int_{\bf B}d(\varphi-\varphi_{o})\wedge \Psi\cr\cr&=& \int_{\bf B}d[(\varphi-\varphi_{o}) \Psi]\cr\cr&= &\int_{\partial {\bf B}}(\varphi-\varphi_{o}) \Psi \end{eqnarray}Now, since $ \frac{1}{p^{*}}+\frac{1}{s}=1$, by H\"older's inequality  we conclude\begin{eqnarray}\int_{\bf B}\Phi\wedge \Psi &<\!\!< &\int_{\partial {\bf B}}|\varphi-\varphi_{o}|\, |\Psi|\,d{\cal H}^{n-1}\cr\cr&<\!\!< &\Big(\int_{\partial {\bf B}}|\varphi-\varphi_{o}|^{p^{*}}d{\cal H}^{n-1}\Big)^{1\over p^{*}} \Big(\int_{\partial {\bf B}}|\Psi|^{s}d{\cal H}^{n-1}\Big)^{1\over s} \cr\cr&<\!\!< &\Big(\int_{\partial {\bf B}}|\Phi|^{p}d{\cal H}^{n-1}\Big)^{1\over p} \Big(\int_{\partial {\bf B}}|\Psi|^{s}d{\cal H}^{n-1}\Big)^{1\over s}\end{eqnarray}\bigskip\bigskipOne can easily deduce that \begin{cor}\label{isop} Let $(\Phi,\Psi)$ be an admissible pair. Given $x\in X$, for almost every   ball  ${\bf B}={\bf B}(x,\rho)$ in $X$ we have\begin{equation}\label{lavagna}\int_{\bf B}\Phi\wedge \Psi<\!\!< \Bigg(\int_{\partial{\bf B}}\Big(|\Phi|^{p}+|\Psi|^{s}\Big) d{\cal H}^{n-1}\Bigg)^{\frac{n}{n-1}} \end{equation}provided $ 1< s=\frac{p(n-1)}{np-n+1}$ and $0<\rho\le R_{X}$.\end{cor}\bigskip\bigskipNext we fix $\omega \in {\cal C}^\infty(\wedge^{n} Y)$ and a Sobolev mapping $f\in {\cal W}^{1,\frac {n^{2}}{n+1}}(X,Y)$.The pointwise Jacobian ${\cal J}_{\omega}(x,f)$ is defined by the rule $${\cal J}_{\omega}(x,f)dx=f^{\sharp}\omega$$Suppose that the target manifold $Y$ has one nontrivial cohomology ${\cal H}^{l}( Y)$,  $1<l<n-1$, then every $n$-form $\omega \in {\cal C}^\infty(\wedge^{n} Y)$ decomposes into wedge products of closed forms of degree smaller than $n$, namely$$\omega= \sum_{i=1}^{N} \a_{i} \wedge \b_{i}\qquad {\rm deg}\alpha_{i}> 1,\qquad {\rm deg}\beta_{i}> 1 $$These forms, once pulled back via the mapping $f:X \to Y$, bring us to analogous decomposition of $f^{\sharp}\omega$:$${\cal J}_{\omega}(x,f)dx=f^{\sharp}\omega= \sum_{i=1}^{N} f^{\sharp} \a_{i} \wedge f^{\sharp} \b_{i}$$see \cite{[4]}.Therefore we obtain the following variant of Theorem \ref{dis}  reminiscent to the classical isoperimetric inequality (\ref{isopuno}).\medskip\begin{cor}\label{discor} Let $f\in {\cal W}^{1,\frac {n^{2}}{n+1}}(X,Y)$ be an orientation  preserving mapping. If   $$\liminf_{t\to \infty}t^{\frac {n+1}{n}}\int_{|Df|>t}|Df|^{\frac {n^{2}}{n+1}}=0$$  then \begin{equation}\int_{\bf B}{\cal J}_{\omega}(x,f)dx<\!\!< \Bigg(\int_{\partial{\bf B}}|Df|^{n-1}d{\cal H}^{n-1}\Bigg)^{\frac{n}{n-1}} \end{equation}for almost every ball  ${\bf B}={\bf B}(x,\rho)$ in $X$ with $0<\rho<R_{X}$.\end{cor}\medskip\noindent {\it Proof:}\qquad We need only consider one wedge product $\omega= \a \wedge \b$ with $\a \in {\cal C}^\infty(\wedge^{l} X)\cap$ ker $d$ and $\b \in {\cal C}^\infty(\wedge^{n-l} X)\cap$ ker $d$.Taking $f^{\sharp}\a$ and $f^{\sharp}\b$ for  the closed forms $\Phi$ and $\Psi$  in Theorem \ref{dis} and  using the pointwise inequalities$$|f^{\sharp}\alpha|<\!\!< |Df|^{l}\qquad\qquad |f^{\sharp}\b|<\!\!< |Df|^{n-l}$$ we get\begin{eqnarray*}\int_{\bf B}{\cal J}_{\omega}(x,f)dx &<\!\!<& \Big(\int_{\partial{\bf B}}|f^{\sharp}\alpha|^{p}d{\cal H}^{n-1}\Big)^{\frac{1}{p}} \Big(\int_{\partial{\bf B}}|f^{\sharp}\b|^{s}d{\cal H}^{n-1}\Big)^{\frac{1}{s}}\cr\cr&<\!\!<& \Big(\int_{\partial{\bf B}}|Df|^{pl}d{\cal H}^{n-1}\Big)^{\frac{1}{p}} \Big(\int_{\partial{\bf B}}|Df|^{s(n-l)}d{\cal H}^{n-1}\Big)^{\frac{1}{s}}\end{eqnarray*}In order to conclude it is  sufficient to choose $ p=\frac{n-1}{l}$ and $s=\frac{n-1}{n-l}$ in previous  inequality. Note that this choice is legitimate thank to assumption $1<l<n-1$.\bigskipTake $n=2l$ and denote by $D^{l\times l}f$ the matrix of all $l\times l$-subdeterminants of $Df$. If ${\cal H}^{l}( Y)\not=0$, the pointwise inequalities $$|f^{\sharp}\alpha|<\!\!< |D^{l\times l}f|\qquad\qquad |f^{\sharp}\b|<\!\!< |D^{l\times l}f|$$hold.The following estimate,  reminiscent to the isoperimetric type inequality (\ref{isopdue}), can be obtained arguing as in the previous corollary.\medskip\begin{cor}\label{scor} Let $f\in {\cal W}^{1,\frac {n^{2}}{n+1}}(X,Y)$ be an orientation  preserving mapping. If   $$\liminf_{t\to \infty}t^{\frac {n+1}{n}}\int_{|Df|>t}|Df|^{\frac {n^{2}}{n+1}}=0$$  then \begin{equation}\int_{\bf B}{\cal J}_{\omega}(x,f)dx<\!\!< \Bigg(\int_{\partial{\bf B}}|D^{\frac{n}{2}\times \frac{n}{2}}f|^{2(n-1)\over n}d{\cal H}^{n-1}\Bigg)^{\frac{n}{n-1}} \end{equation}for almost every ball  ${\bf B}={\bf B}(x,\rho)$ in $X$ with $0<\rho<R_{X}$.\end{cor}\bigskip\bigskip\section{Some applications}\bigskip\bigskipAs an application of the isoperimetric type inequality proved in Theorem \ref{dis}, we give a Morrey's type  estimate   for the wedge product of some special closed forms. Namely we have the following\medskip\begin{thm}\label{is} Let $(\Phi,\Psi)$ be a pair of closed differential forms in ${\cal L}^q(\wedge ^l X)\times{\cal L}^{q'}(\wedge ^{n-l} X)$, where  $1< q,q'<\infty$ are H\"older conjugate exponents. If there exists a constant $K>1$ such that\begin{equation}\label{distorsione}    |\Phi|^{q}+|\Psi|^{q'}\le K (\Phi \wedge\Psi)\end{equation}then \begin{equation}\int_{{\bf B}(0,\rho)}\Phi\wedge \Psi\le \Big(\frac{\rho}{R}\Big)^{\frac{1}{c_{n}K}} \int_{{\bf B}(0,R)}\Phi\wedge \Psi\end{equation}for all  $0<\rho<R =R_{X}$.\end{thm}\medskip\noindent  Remark that inequality (\ref{distorsione}) means that the nonnegative function $\lambda$ such that the $n$-form $\Phi\wedge \Psi=\lambda dx$ satisfies  $|\Phi|^{q}+|\Psi|^{q'}\le K \lambda$.\medskip\noindent {\it Proof:}\quad  Observe that the exponents $p=\frac{q(n-1)}{n}$ and $s=\frac{q'(n-1)}{n}$ are related by the equality $ s=\frac{p(n-1)}{np-n+1}$.The isoperimetric type inequality  (\ref{lavagna}) combined with assumption (\ref{distorsione}) gives\begin{eqnarray}\int_{{\bf B}(0,\rho)}\Phi\wedge \Psi & \le  & c_{n} \Bigg(\int_{\partial {\bf B}(0,\rho)} \Big(|\Phi|^{p}+|\Psi|^{s}\Big) d{\cal H}^{n-1}\Bigg)^{\frac{n}{n-1}}\cr\cr& \le  & c_{n} K \Bigg(\int_{\partial {\bf B}(0,\rho)} (\Phi\wedge \Psi)^{\frac{n-1}{n}}\Bigg)^{\frac{n}{n-1}}\cr\cr& \le  &  c_{n}K \Bigg(\int_{\partial {\bf B}(0,\rho)}dx\Bigg)^{\frac{1}{n-1}} \int_{\partial {\bf B}(0,\rho)}\Phi\wedge \Psi\cr\cr& \le  &  c_{n}K \rho \int_{\partial {\bf B}(0,\rho)}\Phi\wedge \Psi\end{eqnarray}Setting $$h(\rho)=\int_{{\bf B}(0,\rho)}\Phi\wedge \Psi$$a routine calculation shows that $$h(\rho)\le \Big(\frac{\rho}{R}\Big)^{\frac{1}{c_{n}K}} h(R)$$for $\rho\le R$, which concludes the proof.\bigskipNow we deduce the following  H\"older continuity result. \medskip\begin{thm}\label{appli}  Let $(\Phi,\Psi)$ be a pair of closed differential forms in ${\cal L}^q(\wedge ^l X)\times{\cal L}^{q'}(\wedge ^{n-l} X)$, where  $1< q,q'<\infty$ are H\"older conjugate exponents. If there exists a constant $K>1$ such that\begin{equation}    |\Phi|^{q}+|\Psi|^{q'}\le K (\Phi\wedge\Psi)\end{equation}then the (l-1)-form $\varphi$ such that $d\varphi=\Phi$  is H\"older continuous, provided $K<\frac{C_{n}}{n-q}$.\end{thm}    \medskip\noindent {\it Proof}\qquad  Inequality  (\ref{distorsione})  yields$$|d\varphi|^{q}\le K (\Phi\wedge\Psi)$$Then Theorem \ref{is} gives\begin{eqnarray}\int_{{\bf B}(0,\rho)}|d\varphi|^{q} d{\cal H}^{n} & \le  & c_{n} K\int_{ {\bf B}(0,\rho)} \Phi\wedge \Psi\cr\cr& \le  & c_{n} K \Big(\frac{\rho}{R}\Big)^{\frac{1}{c_{n}K}}\int_{{\bf B}(0,R)}\Phi\wedge \Psi\cr\cr & =  & c \rho^{\frac{1}{c_{n}K}}\end{eqnarray}for all  $0<\rho<R =R_{X}$.Therefore $\varphi$  is H\"older continuous of exponent $\gamma=1+\frac{1}{c_{n}K q}-\frac{n}{q}$.  \bigskipNow, following the lead of the classical $q$-harmonic operator, $1<q<\infty$, we consider a bundle map $${\cal H}_{q}: \wedge^{l} X\to \wedge^{l} X $$such that$$|{\cal H}_{q}(\xi)|\le M|\xi|^{q-1}$$$$\langle{\cal H}_{q}(\xi),\xi\rangle \ge M^{-1}|\xi|^{q}$$for every l-form $\xi$ on $X$, where $M\ge 1$ is a constant.Consider a pair $(\Phi,\Psi)\in {\cal L}^q(\wedge^{l} X)\times {\cal L}^{q'}(\wedge^{l} X) $ such that $d\Phi=0$, $d^{*}\Psi=0$ and coupled by\begin{equation}\label{Hodge}    \Psi={\cal H}_{q}(\Phi)\end{equation}    We refer to (\ref{Hodge}) as a Hodge system. Existence and uniqueness  of solutions to (\ref{Hodge}) has been studied in \cite{[5]},  \cite{[6]}. Applying  the operator $d^{*}$ to (\ref{Hodge}) we obtain the second  order elliptic equation  $$d^{*}{\cal H}_{q}(\Phi)=0$$One can easily check that the $l$-form $\Phi$ and the $(n-l)$-form $\Psi$  satisfy inequality (\ref{distorsione}). From Theorem \ref{appli} it follows that the $(l-1)$-form $\varphi$ such that $d\varphi=\Phi$ is H\"older continuous.\bigskip\newpage\begin{thebibliography}{bif}\bibitem{[1]}{\sc M.Carozza, A.Passarelli di Napoli}, {\it Isoperimetric inequality for div-curl fields}, J. of Inequal. and Appl.,vol.7 (3), pp.405--419 (2002)\bibitem{[2]} {\sc H. Cartan},{\it Differential forms},Houghton Mifflin Co., Boston (1970)\bibitem{[3]} {\sc F. Giannetti,T. Iwaniec, J.Onninen, A. Verde},{\it Estimates of Jacobians by subdeterminants} The Journal of Geom. Anal. vol.12 (2), pp.223--254 (2002)\bibitem{[4]}{\sc  P.Hailasz, T.Iwaniec, J.Maly, J.Onninen},{\it Weakly differentiable mappings between manifolds} (2003)\bibitem{[5]} {\sc T.Iwaniec, C.Scott, B.Stroffolini},{\it Nonlinear Hodge theory on manifolds with boundary}, Ann. Mat. Pura Appl.,vol. 4, (1999)\bibitem{[6]} {\sc T.Iwaniec},{\it Nonlinear differential forms}, Lectures in Jyv\"askyl\"a. Report 80. University of Jyv\"askyl\"a, Department of Mathematics (1998)\bibitem{[7]} {\sc C.B.Morrey},{\it Multiple Integrals in the Calculus of Variations}, Springer Verlag, Berlin. (1966)\bibitem{[8]} {\sc S.M\"uller},{\it Higher integrability of determinanats and weak convergence in $L^{1}$}, J. Reine Angew. Math.,vol. 412 , pp.20-34 (1990)\bibitem{[9]} {\sc S.M\"uller, T. Qi, B.S. Yan},{\it On a new class elastic deformations not allowing for cavitation}, Ann. Inst. H. Poincar\'e Anal. Non Linaire,vol. 11 (2), pp.217-243 (1994)\bibitem{[10]} {\sc J.Nash},{\it The imbedding problem for Riemannian manifolds}, Ann. of Math.,vol. 63 (2), pp.20-63 (1956)\bibitem{[11]} {\sc J. Onninen}, {\it A note on the isoperimetric inequality}, Proc. Amer. Math. Soc. vol.131, n.12, pp. 3821-3825 (2003)\bibitem{[12]}{\sc  R.Osserman},{\it The isoperimetric inequality}, Bull. of AMS, vol. 84 (6), pp. 1182--1238 (1978)\bibitem{[13]} {\sc Y.G.Reshetnyak},{\it Mappings with bounded deformation as extremals of Dirichlet type integrals}, Siberian Math. J., vol. 9 , pp. 487--498 (1968)\end{thebibliography}\end{document}