Let $T(f)(x):=\int_{\mathbb{R}^d} k(x,y)f(y) dy$ where $k_1(x)= \int_{\mathbb{R}^d} |k(x,y)|dy$ and $k_2(y)= \int_{\mathbb{R}^d} |k(x,y)|dx$ are two $L^{\infty}$ functions. Then I want to show that $T: L^2 \rightarrow L^2$ has $||T||\le \sqrt{||k_1||_{L^{\infty}}||k_2||_{L^{\infty}}}.$

I was able to show this inequality without the square-root, i.e.

$$||T||_{L^2} = \sup_{||\phi||_2=||\psi||_2=1} \langle T \psi, \phi \rangle$$ $$\le \sup_{||\phi||_2=||\psi||_2=1} \int_{\mathbb{R}^{2d}} |k(x,y)| |\phi(x)| |\psi(y)| dx dy$$ and now by Cauchy schwartz $$\le \sup_{||\phi||_2=||\psi||_2=1} \int_{\mathbb{R}^{2d}} |k(x,y)| |\phi(x)|^2 dx dy\int_{\mathbb{R}^{2d}}|k(x,y)||\psi(y)| dx dy$$ $$\le \sup_{||\phi||_2=||\psi||_2=1} \int_{\mathbb{R}^{d}}|k_1(y)| |\phi(y)|^2 dy \int_{\mathbb{R}^{d}}|k_2(x)| |\phi(x)|^2 dx$$

$$\le ||k_1||_{\infty} ||k_2||_{\infty}$$

Now, I was wondering whether the more inequality stated on the exercise sheet is still correct or whether there has to be some error?


You forgot a square root in the Cauchy-Schwarz inequality. To be more precise: Choose $\phi, \psi \in L^{2}(\mathbb{R}^d)$ with $\lvert \lvert \phi \rvert \rvert_2 = \lvert \lvert \psi \rvert \rvert_2 = 1$. Then \begin{align} \lvert \langle T \psi, \phi \rangle \rvert = \Bigg| \int_{\mathbb{R}^d} \int_{\mathbb{R}^d} k(x,y) \psi(x) \phi(y) \, \textrm{d}y \, \textrm{d}x \Bigg| \leq \int_{\mathbb{R}^d} \int_{\mathbb{R}^d} \big( \lvert k(x,y) \rvert^{1/2} \lvert \phi(x) \rvert \big) \big( \lvert k(x,y) \rvert^{1/2} \lvert \psi(y) \rvert \big) \, \textrm{d}y \, \textrm{d}x. \end{align} Now, we can apply Cauchy-Schwarz' inequality (or Hölder inequality) and Tonelli's theorem, and we obtain that \begin{align} \lvert \langle T \psi, \phi \rangle \rvert &\leq \Bigg( \int_{\mathbb{R}^d} \int_{\mathbb{R}^d} \lvert k(x,y) \rvert \lvert \phi(x) \rvert^{2} \, \textrm{d}y \, \textrm{d}x \Bigg)^{1/2}\Bigg( \int_{\mathbb{R}^d} \int_{\mathbb{R}^d} \lvert k(x,y) \rvert \lvert \psi(y) \rvert^{2} \, \textrm{d}x \, \textrm{d}y \Bigg)^{1/2}\\ & = \Bigg( \int_{\mathbb{R}^d} k_1(x) \lvert \phi(x) \rvert^{2} \, \textrm{d}x \Bigg)^{1/2}\Bigg( \int_{\mathbb{R}^d} k_2(y) \lvert \psi(y) \rvert^{2} \, \textrm{d}x \Bigg)^{1/2}\\ & \leq \Big( \lvert \lvert k_1 \rvert \rvert_{L^{\infty}(\mathbb{R}^d)} \lvert \lvert k_2 \rvert \rvert_{L^{\infty}(\mathbb{R}^d) } \lvert \lvert \phi \rvert \rvert_2 \lvert \lvert \psi \rvert \rvert_2 \Big)^{1/2} = \sqrt{ \lvert \lvert k_1 \rvert \rvert_{L^{\infty}(\mathbb{R}^d)} \lvert \lvert k_2 \rvert \rvert_{L^{\infty}(\mathbb{R}^d) } }. \end{align} As this is true for all $\phi, \psi \in L^{2}(\mathbb{R}^d)$ with $\lvert \lvert \phi \rvert \rvert_2 = \lvert \lvert \psi \rvert \rvert_2 = 1$ the claim follows.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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