This note is taken in the PDE discussion group in 2012, on the topic of important spaces in fluid dynamics.
Lecture 2: Carleson measure
- Non-tangential maximal functions
- Unit cone at point $x$: $\Gamma(x):=\{(y,t)\in\mathbb{R}^n\times\mathbb{R}^+~:~|y-x|<t\}$.
- Non-tangential MF: $\displaystyle F^*(x)=\sup_{(y,t)\in\Gamma(x)}|F(y,t)|$.
- Carleson measure
- Tent mapping on balls: $T:\text{Ball}(\mathbb{R}^n)\to\text{Cylinder}(\mathbb{R}^n\times\mathbb{R}^+)$,
\[T(B(x,r))=\{(y,t)~:~y\in B(x,r), t\in[0,2r]\}.\] - Tent mapping on cubes: $T:\text{Cube}(\mathbb{R}^n)\to\text{Cube}(\mathbb{R}^n\times\mathbb{R}^+)$, $Q(x,l)$ is a cube centered at $x$, side length $l$.
\[T(Q(x,l))=\{(y,t)~:~y\in B(x,r), t\in[0,l]\}.\]
Definition (Carleson measure): A positive measure $\mu$ is called a Carleson measure if
\[\|\mu\|_C:=\sup_Q\frac{1}{|Q|}\mu(T(Q))<\infty.\]
- $\displaystyle C(\mu)(x):=\sup_{\{Q:x\in Q\}}\frac{1}{|Q|}\mu(T(Q))$. Note that $\|\mu\|_C=\sup_x C(\mu)(x)$.
- Whitney decomposition
Let $\Omega$ be an open, nonempty, proper subset of $\mathbb{R}^n$. There exists a family of closed cubes $\{Q_j\}_j$ such that
- $\cup_j Q_j=\Omega$, $Q_j$'s have disjoint interiors.
- $\sqrt{n}~l(Q_j)\leq dist(Q_j,\Omega^c)\leq4\sqrt{n}~l(Q_j)$.
- If the boundaries of $Q_j$ and $Q_k$ are touched, then $\displaystyle\frac{1}{4}\leq\frac{l(Q_j)}{l(Q_k)}\leq4$.
- Given $Q_j$, there exists at most $12^n$ $Q^k$'s that touch it.
- Carleson measure in $\mathbb{R}^{n+1}$ can be controlled by $C(\mu)$ in $\mathbb{R}^n$.
Theorem 1: There exists a constant $C_n$ such that for any $\alpha>0$, any positive measure $\mu$ on $\mathbb{R}_+^{n+1}$ and any $\mu$-measurable function $F$ on $\mathbb{R}_+^{n+1}$, we have
\[\mu\left(\{(x,t)\in\mathbb{R}_+^{n+1}:|F(x,t)|\geq\alpha\}\right)\leq C_n\int_{\{F^*\geq\alpha\}}C(\mu)(x)dx.\]
Proof: Let $\Omega=\{F^*>\alpha\}$ be an open set in $\mathbb{R}^n$.
Step 1: Prove $\{|F|>\alpha\}\subset\cup_{z\in\Omega}T(B(z,\delta(z))$, where $\delta(z)=dist(z,\Omega^c)$.
Suppose $(x,t)\in\{|F|>\alpha\}$. As $(x,t)$ lies inside the cone of any point inside $B(x,t)$, we have $B(x,t)\subset\Omega$. Then, $t\leq\delta(z)$, which implies $(x,t)\in T(B(x,\delta(x))$.
Step 2: Prove $B(z,\delta(z))\subset 12B_k$, where $B_k$ is the smallest ball containing $Q_k$, and $Q_k$ is one of the cube from Whitney decomposition which contains $z$.
$\delta(z)\leq\sqrt{n}~l(Q_k)+dist(Q_k,\Omega^c)\leq5\sqrt{n}~l(Q_k)$.
For any point $y\in B(z,\delta(z))$, the distance between $y$ and the center of $Q_k$ is bounded by $\delta(z)+\sqrt{n}~l(Q_k)\leq6\sqrt{n}~l(Q_k)$. As the radius of $B_k$ is $(\sqrt{n}/2)l(Q_k)$, $y\in 12B_k$.
Step 3: From the first two steps, we get: LHS$\leq\sum_k\mu(T(12B_k))\leq\sum_k\mu(T(12\sqrt{n}Q_k))$.
By definition of $C(\mu)$, $\mu(T(12\sqrt{n}Q_k))\leq |12\sqrt{n}Q_k|C(\mu)(x)$, for all $x\in Q_k$. Finally,
\[\text{LHS}\leq\sum_k|12\sqrt{n}Q_k|\inf_{x\in Q_k}C(\mu)(x)\leq\sum_k\frac{|12\sqrt{n}Q_k|}{|Q_k|}\int_{Q_k}C(\mu)(x)dx=(12\sqrt{n})^n\int_\Omega C(\mu)(x)dx.\]
Corollary 1 ($L^p$ Control): $\displaystyle \int_{\mathbb{R}_+^{n+1}}|F(x,t)|^pd\mu(x,t)\leq C_n\|\mu\|_C\int_{\mathbb{R}^n}(F^*(x))^pdx$.
Proof:
\begin{align*}LHS=&\int_0^\infty\alpha^p\mu(|F(x,t)|\geq\alpha|) d\alpha\leq C_n\int_0^\infty\alpha^p\int_{F^*\geq\alpha}C(\mu)(x)dxd\alpha\\ \leq&C_n\|\mu\|_C\int_0^\infty\alpha^p|\{F^*\geq\alpha\}|d\alpha=RHS.\end{align*}
- Carleson measure generated by a BMO function
Theorem 2: Let $b$ be a BMO function on $\mathbb{R}^n$ and let $\psi$ be an integrable function with mean value zero on $\mathbb{R}^n$ that satisfies
\[|\psi(x)|\leq A(1+|x|)^{-n-1},\qquad\sup_{\xi\in\mathbb{R}^n}\sum_j|\hat{\psi}(2^{-j}\xi)|^2\leq A^2<\infty.\]
Then, there exists a constant $C_n$ such that
\[d\mu(x,t)=\sum_{j\in\mathbb{Z}}|\Delta_j(b)|^2dx\cdot\delta_{2^{-j}}(t)\]
is a Carleson measure on $\mathbb{R}_+^{n+1}$ with norm at most $C_n A^2\|b\|_{BMO}^2$. Here, $\Delta_j$ is the Littlewood-Paley operator $\Delta_j(b)=b\star\psi_{2^{-j}}$, and $\delta$ is the Dirac mass.
Proof: We are going to show that, $\mu(T(Q))\leq|Q|C_nA^2\|b\|_{BMO}^2$, for all cubes $Q\subset\mathbb{R}^n$.
By definition $\mu(T(Q))=\sum_{2^{-j}\leq l(Q)}\int_Q|\Delta_j(b)(x)|^2dx$.
Let $Q^*$ be a larger cube with the same center as $Q$ and side length $3\sqrt{n}~l(Q)$. Decompose $b$ into 3 parts.
\[b=(b-b_Q)\cdot\textbf{1}_{Q^*}+(b-b_Q)\cdot\textbf{1}_{(Q^*)^c}+b_Q.\]
Then, we have
\[|\Delta_j(b)|^2\leq3\big[|\Delta_j((b-b_Q)\cdot\textbf{1}_{Q^*})|^2+\Delta_j((b-b_Q)\cdot\textbf{1}_{(Q^*)^c})|^2+\Delta_j(b_Q)|^2\big].\]
The third part $\Delta_j(b_Q)=0$ as $\psi$ has mean zero.
Now, we estimate the first term $\Sigma_1=\sum_{j\in\mathbb{Z}}\int_{\mathbb{R}^n}|\Delta_j((b-b_Q)\cdot\textbf{1}_{Q^*})|^2dx$.
\begin{align*}\Sigma_1\leq\sum_{j\in\mathbb{Z}}\int_{\mathbb{R}^n}|\hat{\psi}(2^{-j}\xi)|^2|\widehat{(b-b_Q)\cdot\textbf{1}_{Q^*}}|^2d\xi\leq A^2\int_{Q^*}|b-b_Q|^2dx.\end{align*}
Here, $\int_{Q^*}|b-b_Q|^2\leq\int_{Q^*}|b-b_{Q^*}|^2+|Q^*||b_{Q}-b_{Q^*}|^2\lesssim_n |Q^*||b|_{BMO}^2 \lesssim_n |Q|\|b\|_{BMO}^2$.
We are left with the estimate of $\Sigma_2=\sum_{2^{-j}\leq l(Q)}\int_Q|\Delta_j((b-b_Q)\cdot\textbf{1}_{(Q^*)^c})|^2dx$. As $|\psi(x)|\leq A(1+|x|)^{-n-1}$, we get $|\psi_{2^{-j}}(x)|\leq2^{jn}A(1+|2^jx|)^{n+1}$.
\begin{align*}\Sigma_2\leq&\sum_{2^{-j}\leq l(Q)}\int_Q\left|\int_{(Q^*)^c}\frac{A2^{jn}|b(y)-b_Q|}{(1+|2^j(x-y)|)^{n+1}}dy\right|^2dx\\ \leq&\sum_{2^{-j}\leq l(Q)}2^{-2j}\int_Q\left|\int_{(Q^*)^c}\frac{A|b(y)-b_Q|}{(2^{-j}+|(x-y)|)^{n+1}}dy\right|^2dx\\ \leq& \frac{4}{3}l(Q)^2\int_Q\left|\int_{(Q^*)^c}\frac{A|b(y)-b_Q|}{(l(Q)+|(x-y)|)^{n+1}}dy\right|^2dx.\end{align*}
To get a bound of the inside integral, we decompose $(Q^*)^c$ to ring in order to get a better bound for the denominator. $\{2Q^*\backslash Q^*,\cdots,2^kQ^*\backslash 2^{k-1}Q^*,\cdots\}$.
\[\int_{2^kQ^*\backslash2^{k-1}Q^*}\frac{A|b(y)-b_Q|}{(l(Q)+|(x-y)|)^{n+1}}dy\leq\left(\frac{l(Q)}{2}+2^{k-1}l(Q^*)\right)^{-n-1}\int_{2^kQ^*}|b(y)-b(Q)|dy.\]
Here, $\int_{2^kQ^*}|b-b_Q|\leq A\int_{2^kQ^*}|b-b_{2^kQ^*}|+|2^kQ^*||b_{Q}-b_{2^kQ^*}|\lesssim_n k2^{kn}|Q|\|b\|_{BMO}$. Therefore, we have
\[\int\star~dy\lesssim_n A2^{-(k-1)(n+1)}l(Q)^{-n-1}\cdot k2^{kn}|Q|\|b\|_{BMO}=Ak2^{n+1-k}l(Q)^{-1}\|b\|_{BMO}.\]
Sum on $k$ yields a finite number. Plug into $\Sigma_2$, the $l(Q)$ term is cancelled. It gives the desired estimate.