Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

let $1 = \frac{1}{p} + \frac{1}{q}$ as usual. Let $f \in L^p, g \in L^q$ be vector fields from $\mathbb R^n$ to itself.

Assume $div f = 0$ and there exists a function $G$ s.t. $\nabla G = g$. Then $f \cdot g \in \mathcal H^1$ is a Hardy space function.

Do you know where I can find a proof of this conclusion? I am aware of a paper by Coifman et al. "Compensated compactness and Hardy spaces", but I am not granted access to this journal. Hence I am looking for an alternative resource.

share|cite|improve this question
    
Do you have any sort of library available, or are looking for an online reference? What books have you tried? I bet one of Evans' "Partial Differential Equations," Grafakos' "Classical Fourier Analysis," or Stein's "Harmonic Analysis" have a proof. – JavaMan Mar 29 '11 at 15:49
    
Hi @Martin, did you end up looking through those books? – Glen Wheeler Apr 14 '11 at 7:35
    
you find it in the book "weak convergence methods...." by evans – user34172 Jun 21 '12 at 14:09
    
arxiv.org/abs/0712.2133 may be relevant. – Willie Wong Jun 21 '12 at 16:08

A remark on the div-curl lemma by P. G. Lemarié-Rieusset gives a complete proof of the result in greater generality, with $L^p-L^q$ replaced by a Calderón-Zygmund pair of Banach spaces.

As pointed out in comments, book presentation is also available: Weak convergence methods for nonlinear partial differential equations by L. C. Evans.

share|cite|improve this answer
    
Unless I'm looking in the wrong place, the book by Evans does not show that $f\cdot g\in\mathcal{H}^{1}(\mathbb{R}^{n})$ in its treatment of the div-curl lemma. I was unable to find any mention of Hardy spaces. – Matt Rosenzweig Jan 1 at 1:48

Since the original paper by Coifman, Lions, Meyer, and Semmes is not easily accessible--I had to physically scan the original journal--I thought I would include in this thread the simplest proof of the result in question. Let $E\in L^{p}(\mathbb{R}^{n}\rightarrow\mathbb{R}^{n})$ and $B\in L^{p'}(\mathbb{R}^{n}\rightarrow\mathbb{R}^{n})$, where $\frac{1}{p}+\frac{1}{p'}=1$ and $1<p<\infty$, be div- and curl-free vector fields respectively.

Lemma. Let $E$ and $B$ be as above. Let $h\in C_{c}^{\infty}(\mathbb{R}^{n})$ (wlog supported in unit ball). For all $\alpha,\beta$ such that $$\dfrac{1}{\alpha}+\dfrac{1}{\beta}=1+\dfrac{1}{n}, \quad 1\leq \alpha\leq p, \enspace 1<\beta\leq p'$$ there exists a constant $C=C(\alpha,\beta)>0$ such that $$|\left(h_{t}\ast (E\cdot B)\right)(x)|\leq C\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|E|^{\alpha}\right)^{1/\alpha}\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|B|^{\beta}\right)^{1/\beta},$$ $\forall x\in\mathbb{R}^{n}, t>0$.

Proof of Lemma. By density, we may assume briefly that $B$ is Schwartz; this hypothesis is just to justify the following manipulations. Since $B$ is curl-free, a little Fourier analysis ($n\geq 2$) shows that $$\widehat{B}=\xi(\dfrac{\xi}{|\xi|^{2}}\cdot\widehat{B})=\widehat{\nabla\phi},\quad \widehat{\phi}:=\dfrac{\xi}{(2\pi i)|\xi|^{2}}\cdot\widehat{B}$$ in the sense of tempered distributions.

I claim that $\phi\in L^{p*}(\mathbb{R}^{n})$, if $1<p'<n$, and $\phi\in L_{loc}^{q}(\mathbb{R}^{n})$ for all $q<\infty$, if $p'\geq n$. Indeed, $\frac{\xi}{|\xi|}$ is, up to a multiplicative constant, the symbol of the vector Riesz transform hence bounded on $L^{r}(\mathbb{R}^{n}\rightarrow\mathbb{R}^{n})$, for all $1<r<\infty$. $|\xi|^{-1}$ is, up to a multiplicative constant, the Fourier transform of $|x|^{-n+1}$. The claim then follows from the Hardy-Littlewood-Sobolev lemma. If $n=1$, then just use FTC.

Next, I claim that $E\cdot B=\text{div}(E\phi)$ in $\mathcal{D}'$. Formally, this is obvious (just use product rule and the hypothesis that $\text{div}(E)=0$) but to prove this rigorously we use mollification. Let $\psi$ be a $C_{c}^{\infty}$ mollifier. Then \begin{align*} \text{div}((E\ast\psi_{\varepsilon})(\psi_{\varepsilon}\ast\phi))&=\text{div}(E\ast\psi_{\varepsilon})(\psi_{\varepsilon}\ast\phi)+(E\ast\psi_{\varepsilon})\cdot ((\nabla\phi)\ast\psi_{\varepsilon})\\ &=\text{div}(E)\ast\psi_{\varepsilon}+(E\ast\psi_{\varepsilon})\cdot(B\ast\psi_{\varepsilon})\\ &=(E\ast\psi_{\varepsilon})\cdot(B\ast\psi_{\varepsilon})\\ \end{align*} Since $E\ast\psi_{\varepsilon}\rightarrow E$ in $L^{p}$ and $B\ast\psi_{\varepsilon}\rightarrow B$ in $L^{p'}$, the last expression above $\rightarrow E\cdot B$, as $\varepsilon\rightarrow 0$, in $L^{1}$, hence in distribution. Similarly, the LHS $\rightarrow\text{div}(E\phi)$ in $\mathcal{D}'$.

So, we may write \begin{align*} (h_{t}\ast(E\cdot B))(x)&=\int_{\mathbb{R}^{n}}(\nabla h)(\frac{x-y}{t})t^{-n-1}\cdot E(y)\phi(y)dy\\ &=\int_{\mathbb{R}^{n}}(\nabla h)(\frac{x-y}{t})t^{-n-1}\cdot E(y)\left[\phi(y)-\frac{1}{|B_{t}(x)|}\int_{B_{t}(x)}\phi\right]dy \end{align*} where the last line follows by the hypothesis that $\text{div}(E)=0$ and $|B_{t}|^{-1}\int_{B_{t}(x)}\phi$ is a constant. Using H\"{o}lder's inequality with $\beta$, we obtain \begin{align*} |(h_{t}\ast (E\cdot B))(x)|\lesssim_{n}\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|E|^{\beta}dy\right)^{1/\beta}\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}\left|[\phi-\frac{1}{|B_{t}(x)|}\int_{B_{t}(x)}\phi]t^{-1}\right|^{\beta'}dy\right)^{1/\beta'} \end{align*} Using the Poincare inequality, we can estimate the second factor by $$\lesssim_{n,\beta,\alpha}\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|\nabla \phi|^{\alpha}\right)^{1/\alpha}$$ since $\frac{1}{\alpha}-\frac{1}{n}=1-\frac{1}{\beta}=\frac{1}{\beta'}$ (i.e. $\beta'$ is the Sobolev conjugate of $\alpha$). Recalling that $\nabla\phi=B$ completes the proof of the lemma. $\Box$

Proof of Main Resuklt. We use the maximal function characterization of Hardy spaces with $h$ as our Schwartz function to show that $E\cdot B\in\mathcal{H}^{1}(\mathbb{R}^{n})$. Since $\frac{1}{p}+\frac{1}{p'}=1$, we can find $\alpha<p$, $\beta<p'$ as in the statement of the lemma. Whence, \begin{align*} \sup_{t>0}\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|E|^{\alpha}\right)^{1/\alpha}\left(\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|B|^{\beta}\right)^{1/\beta}&\leq\left(\sup_{t>0}\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|E|^{\alpha}\right)^{1/\alpha}\left(\sup_{t>0}\dfrac{1}{|B_{t}(x)|}\int_{B_{t}(x)}|B|^{\beta}\right)^{1/\beta}\\ &\leq (M(|E|^{\alpha})(x))^{1/\alpha}(M(|B|^{\beta})(x))^{1/\beta}, \end{align*} where $M$ denotes the Hardy-Littlewood maximal function. Observe that $|E|^{\alpha}\in L^{p/\alpha}(\mathbb{R}^{n})$ and $|B|^{\beta}\in L^{p'/\beta}(\mathbb{R}^{n})$, where $p/\alpha, p'/\beta>1$. Since $M$ is bounded on $L^{r}(\mathbb{R}^{n})$, for $1<r\leq\infty$, we conclude from Holder's inequality that \begin{align*} \int_{\mathbb{R}^{n}}(M(|E|^{\alpha}))^{1/\alpha}(M(|B|^{\beta}))^{1/\beta}dx&\leq\|(M(|E|^{\alpha}))^{1/\alpha}\|_{L^{p}}\|(M(|B|^{\beta}))^{1/\beta}\|_{L^{p'}}\\ &=\|M(|E|^{\alpha})\|_{L^{p/\alpha}}^{1/\alpha}\|M(|B|^{\beta})\|_{L^{p'/\beta}}\\ &\lesssim_{n,p,\alpha}\|E\|_{L^{p}}\|B\|_{L^{p'}}, \end{align*} which completes the proof.

share|cite|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.