1
$\begingroup$

For some numerical analysis of a fluid, I am wondering if there is any inequality that provides a lower bound (in the $L^2$ norm), for the $L^2$ inner product of a quantity with its derivative. In particular, I am seeking one that is useful for the form: $$(D(x),D(v))_{L^2} + (\nabla\cdot{x}, \nabla\cdot{v})_{L^2} + (\nabla\cdot{v}, \nabla\cdot{v})_{L^2}\ge C\|\nabla x\|^2_{L^2} + K\|\nabla v\|^2_{L^2}$$ for constants $C > 0$ and $K \ge 0$, where strain-rate tensor $D(v)\equiv \frac12(\nabla v + (\nabla v)^T)$, $t$ represents time, and velocity $v = \frac{d}{dt}x$ for displacement $x$.

I am trying to see if I can show that the left hand side is coercive in terms of $\nabla x$. i.e. it is okay if $K$ may be zero.

$x$ and $v$ are in the Sobolev space $V=\{\varphi \in (H^1(\Omega))^d \;|\; \varphi = 0\textrm{ on } \partial\Omega\}$ where $\Omega$ is a $d-$dimensional region of the fluid flow, for $d=2,3$.

P.S. If the inequality were to be reversed, I would have used the fact that $\|\nabla\cdot{v}\|_{L^2} \le \sqrt{d}\|\nabla{v}\|_{L^2}$ for a $d$-dimensional domain; and also applied Cauchy-Schwartz inequality, then Young's inequality, and Korn's inequality.

This answer on Math SE mentions Kantorovich inequalities, but there is no reference and my searches so far have been futile.

$\endgroup$
  • 1
    $\begingroup$ What are the domains of $x$ and $v$? $\endgroup$ – user126920 Sep 17 '18 at 1:06
  • $\begingroup$ Thank you, I have added more context to the question! $\endgroup$ – Cogicero Sep 17 '18 at 1:23
2
$\begingroup$

I don't think what you want is possible.

In phase space $x$ and $v$ are basically quantities that can be independently prescribed. The first terms of your left hand side is linear in $v$. So fixing $x$, $v$ arbitrary, for $\lambda$ of the correct sign and $|\lambda|$ sufficiently small, you will have that the left hand side is negative if you replace $v$ by $\lambda v$. A negative quantity can never be coercive!

$\endgroup$
  • $\begingroup$ Thanks for your useful answer! I am inclined to accept this right away and move on, but a friend just suggested I could try integrating by parts on the first term on the left. Does this look like it makes sense? I'm clutching at straws here. Thanks! $\endgroup$ – Cogicero Sep 17 '18 at 23:02
  • $\begingroup$ Also, I am sorry I see how the left is linear in $x$ so I think that infers linearity in $v$. However I can't see how the left hand side is negative. Could you please expand on that? Thanks! $\endgroup$ – Cogicero Sep 18 '18 at 5:36
  • 1
    $\begingroup$ Essentially, you have some vector $A$ (which I use as a stand in for $\nabla x$) and some vector $B$ (which I use as a stand in for $\nabla v$). Your left hand side is basically $\langle A,B\rangle + \langle B, B\rangle$. Let $B' = \lambda B$. As long as $A$ and $B$ is not orthogonal, then you have $$ \langle A, B'\rangle + \langle B', B'\rangle = \lambda (\langle A, B\rangle + \lambda \langle B, B\rangle) $$ so for any $\lambda$ between $0$ and $-\langle A,B\rangle / \langle B, B\rangle$ the quantity is negative. $\endgroup$ – Willie Wong Sep 18 '18 at 13:59
  • $\begingroup$ Thanks again. This is clear now. I have accepted the answer, but please I have one more related question. Does it change anything if the $\langle B,B \rangle$ term is not on the left i.e. the left hand is simply $(D(x),D(v))_{L^2} + (∇⋅x,∇⋅v)_{L^2}$ (that's what I get when I approach the problem a different way)? It seems to me, that it would mean $\langle A,B \rangle$ is on the left, which isn't necessarily non-negative. $\endgroup$ – Cogicero Sep 18 '18 at 16:29
  • 1
    $\begingroup$ Without the $\langle B, B\rangle$ term the situation is worse. You don't even need $\lambda$ sufficiently small. If it has the "correct" sign it breaks the coercivity. $\endgroup$ – Willie Wong Sep 18 '18 at 16:55

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.