I posted this on Math StackExchange, but I figured it couldn't hurt to ask here as well.

I'm trying to decipher a particular claim in a paper I'm reading, but I just can't seem to figure it out.

The M. Riesz Interpolation Theorem says:

Let $T:L^{p_0}\cap L^{p_1} \to L^{q_0} \cap L^{q_1}$ be a linear map satisfying: $$||Tf||_{q_j}\leq M_{j}||f||_{p_j}, \quad j=0,1$$ with $1\leq p_j,q_j\leq\infty$. Then, if for $0< t <1$, we define $$\frac{1}{p_t}=\frac{1-t}{p_0}+\frac{t}{p_1}, \quad \frac{1}{q_t}=\frac{1-t}{q_0}+\frac{t}{q_1},$$ we have $$||Tf||_{q_t}\leq (M_0)^{1-t}(M_1)^{t}||f||_{p_t}, \quad f\in L^{p_0}\cap L^{p_1}.$$

Note that $p_0 \leq p_t \leq p_1$ and $q_0 \leq q_t \leq q_1$.

The paper I'm reading gives the following:

$$||e_\lambda||_p \leq C \lambda^{\delta(p)}||e_\lambda||_{2}, \quad 2\leq p \leq 6$$

$$\delta(p) = \frac{1}{2}\left(\frac{1}{2} - \frac{1}{p} \right)$$

Here $e_\lambda$ is an eigenfunction (with eigenvalue $\lambda$) of $\sqrt{-\Delta}$ on a fixed, compact Riemannian manifold $M$, with no boundary, and $C$ is a fixed constant depending only on the manifold. Note in particular that $\delta(2)=0$ and $\delta(6)=\tfrac{1}{6}$.

It then proceeds to claim that using the above result for $p=2$ and $p=6$, along with the interpolation theorem, we have the following:

$$\limsup_{k\to\infty} \lambda_{k}^{-\delta(p)}||e_{\lambda_k}||_{L^{p}(M)} = 0$$ is true for $2<p<6$ if and only if it is true for $p=4$, where the eigenfunctions are all taken to have unit $L^{2}(M)$-norms, and $\lambda_1 \leq \lambda_2 \leq \cdots.$

If we let $T=E_{k}$ be the projection onto the $k^{\text{th}}$ eigenspace, that is, $$E_{k}f = \langle f,e_{\lambda_k}\rangle e_{\lambda_k}(x), \quad \langle f,e_{\lambda_k}\rangle = \int_{M}f(x)\overline{e_{\lambda_k}(x)}dx, \quad \forall f\in L^{2}(M)$$ then $E_{k}:L^{2} \to L^{q_0}\cap L^{q_1}$, for any $2\leq q_0,q_1\leq 6$, and it seems like we should be able to use the interpolation theorem in some manner.

I'm lost though when it comes to showing the above claim is true for $2<p<6$, assuming it is true for $p=4$. Moreover, I feel like I've seen this kind of trick a lot, where we only have to prove a statement for some particular value of $p$ in order for it to be true for some larger range of values. What am I missing?


My answer is for $p \in (2,4)$, the other case should follow similar.

Let $t \in (0,1)$ be given, such that $$\frac1p = \frac{1-t}{2}+\frac{t}{4}.$$ I will go to use the following consequence of Hölder (this is similar to the Riesz Interpolation Theorem, but somewhat simpler to use in your case): $$\lVert f \rVert_p \le \lVert f\rVert_2^{1-t} \lVert f\rVert_4^t$$ Using $\delta(p) = (1-t)\;\delta(2) + t \; \delta(4)$ you find $$\lambda_k^{-\delta(p)} \lVert e_k \rVert_p \le \Big(\lambda_k^{-\delta(2)}\lVert e_k\rVert_2\Big)^{1-t}\Big(\lambda_k^{-\delta(4)}\lVert e_k\rVert_4\Big)^t = \Big(\lambda_k^{-\delta(4)}\lVert e_k\rVert_4\Big)^t.$$ This shows, that your lim-sup holds for all $p \in (2,4)$ if it holds for $p = 4$.

Edit: A intuition for the statement is the following: You have $$\lambda_k^{-\delta(p)} \lVert e_k \rVert_p$$ is bounded for $p \in [2,6]$ as $k \to \infty$. If for some $p \in (2,6)$, the $\limsup$ is zero, than you can interpolate and obtain a zero $\limsup$ for all $p \in (2,6)$. Note there is nothing special about $p = 4$.

A similar argument is: if a sequence of functions $f_k$ is bounded in $L^p$ and converges towards zero in $L^q$, $q < p$, then it converges to $0$ in $L^r$ for all $r \in [q,p)$.

  • $\begingroup$ Thanks. That really helps. This is the [Hölder interpolation][1] theorem you're referencing, correct? In the case when $p_t\in(4,6)$ though, we end up with $$\lambda_k^{-\delta(p_t)}||e_{\lambda_k}||_{p_t} \leq \left( \lambda_k^{-\delta(4)}||e_{\lambda_k}||_4 \right)^{1-t}\left( \lambda_k^{-\delta(6)}||e_{\lambda_k}||_6 \right)^{t}$$ and we can no longer rely on the unit-norm in $L^{2}(M)$ nor the fact that $\delta(2)=0$. How do we proceed from here? [1]: en.wikipedia.org/wiki/… $\endgroup$ – Patch Mar 28 '13 at 10:19
  • $\begingroup$ You can use $\lVert e_k \rVert_6 \le C \, \lambda_k^{\delta(6)} \, \lVert e_k \rVert_2$. $\endgroup$ – gerw Mar 28 '13 at 10:41
  • $\begingroup$ Oh, of course! Sorry. Thanks again. $\endgroup$ – Patch Mar 28 '13 at 10:44
  • $\begingroup$ The value $p=4$ might not be special, but it is the only instance in which we are able to prove the $\limsup$ convergence explicitly. $\endgroup$ – Patch Mar 28 '13 at 10:49

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.