In Moser's famous paper harnack inequality for parabolic equations, he used the following simple Poincare inequality(Lemma 3)

$\int (f(x)-k)^2 w(x) dx \leq c(w) \int |\nabla f|^2 w(x) dx$

where $k=\int f(x)w(x)dx / \int w(x)dx$. The weight function $w(x)$ is just a smooth axially decreasing cut-off function of compact support.

Is this still true when LHS is replaced by a higher norm, i.e.

$\int (f(x)-k)^p w(x) dx \leq c(p,w) (\int |\nabla f|^2 w(x) dx)^\frac{p}{2} $ ?

(Of course $p$ should at least satisfy $p < 2^*=\frac{2n}{n-2}$.)

References for weighted sobolev spaces often deal with complicated weight functions(for example satisfying the Muckenhoupt $A_p$-condition). I can't find a clear answer to my question concerning just smooth cut-off weights while I need this inequality when trying to apply Moser iteration for drift-diffusion equations.


This inequality is not true. (For $2<p<2^*$, that is. This is how I understood the question.) To see why, imagine that the function $f$ has a huge peak in the area where the weight $w$ is very small.

To simplify the matters, assume that $w$ looks similar to (smoothed) "stairs" with infinite number of steps of width $\delta_i\,(1\le i<\infty)$. Now, let $f$ be a usual (spherically symmetric) cap with support entirely on the $i$-th "step". One can see that for the above inequality to hold in this case we must have the inequality $$\delta^{n(1-\frac{p}{2^*})}\le Cw^{\frac{p}{2}-1},$$ where $w=w_i$ is the "height" of this "step". Now, for $2<p<2^*$ it is not a problem to design "stairs" where this inequality does not hold. (I presume that the inequality isn't true for any reasonable weight $w$, but to prove this in general may be a little more difficult.)

  • $\begingroup$ Thanks for your explanation! Now I see that the unbalanced w's on two sides of my inequality can lead to trouble. $\endgroup$ – xiao Jan 31 '18 at 20:35

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.