The inequality is the following:

For any $\varepsilon >0$ there exists $C_\varepsilon >0$ s.t.:

$$\lVert u\rVert_{\infty}\leq \varepsilon\ \lVert u^\prime \rVert_{\infty}+C_{\varepsilon}\ \lVert u\rVert_1$$

for any $u\in C^1([0,1])$.

Does anyone know the trick to obtain this inequality? I am pretty sure it is just a little trick.

The constant $C_\varepsilon$ should be something like $\frac{K}{\varepsilon}$ with $K>0$.

  • 2
    $\begingroup$ What is $u'$ here? $\endgroup$ – Martin Argerami Mar 9 '12 at 1:19
  • $\begingroup$ Typo, $u'$ is continuous. $\endgroup$ – checkmath Mar 9 '12 at 1:37
  • $\begingroup$ What is «a Cauchy»? $\endgroup$ – Mariano Suárez-Álvarez Mar 9 '12 at 1:42
  • 4
    $\begingroup$ Hey Downvoters could you just explain your point? $\endgroup$ – checkmath Mar 9 '12 at 2:55
  • 1
    $\begingroup$ Perhaps you should explain the context of this problem a bit more. $\endgroup$ – user642796 Mar 9 '12 at 15:47

The following is not a complete answer, but (hopefully) it could be useful.

It is easy to get the inequality without $\varepsilon$.

In fact, let $u\in C^1([0,1])$. Since $u^\prime$ is bounded (by Weierstrass), $u$ is Lipschitz and: $$|u(x)-u(y)|\leq \lVert u^\prime \rVert_\infty\ |x-y|\; ;$$ by reverse triangle inequality one gets: $$|u(x)|\leq \lVert u^\prime \rVert_\infty\ |x-y| + |u(y)|$$ and an integration w.r.t. $y\in [0,1]$ yields: $$|u(x)|\leq \lVert u^\prime \rVert_\infty\ \left( x^2-x+\frac{1}{2}\right) +\lVert u\rVert_1\; .$$ Maximizing both LH and RH sides w.r.t. $x\in [0,1]$ one obtains: $$\tag{1} \lVert u\rVert_\infty \leq \frac{1}{2}\ \lVert u^\prime \rVert_\infty +\lVert u\rVert_1\; .$$

Therefore a comparison of (1) with your guess for $C_\varepsilon$, i.e. $C_\varepsilon = K/\varepsilon$, yields $C_\varepsilon = \frac{1}{2\varepsilon}$.

Coming to the inequality with $\varepsilon$, I have to think about it...


Suppose first that there is some $t_0$ with $u(t_0) = 0$. Then by the fundamental theorem of calculus, we have $$u(t)^2 = 2 \int_{t_0}^t u'(t) u(t)\,dt \le 2 ||u'||_{\infty} ||u||_1.$$ Thus $$|u(t)| \le \sqrt{(2 \epsilon ||u'||_\infty)(\frac{1}{\epsilon} ||u||_1)}$$ So taking the supremum over $t$ and using the AM-GM inequality we get $$||u||_\infty \le \epsilon ||u'||_\infty + \frac{1}{2\epsilon} ||u||_1.$$

Now suppose there is no $t_0$ with $u(t_0) = 0$. By the intermediate value theorem either $u>0$ everywhere or $u<0$ everywhere. By replacing $u$ by $-u$ if necessary we assume $u > 0$ everywhere. Let $t_1$ be the point where $u$ attains its minimum. Set $\tilde{u}(t) = u(t) - u(t_1)$. Note that $\tilde{u}' = u'$ and $||\tilde{u}||_1 = ||u||_1 - u(t_1) \le ||u||_1$. Applying the previous case to $\tilde{u}$ we have $$||u||_\infty = ||\tilde{u}||_\infty + u(t_1) \le \epsilon ||u||_\infty + \frac{1}{2\epsilon} ||u||_1 + u(t_1).$$ Finally, by integrating the inequality $u(t_1) \le u(t) = |u(t)|$, we have $u(t_1) \le ||u||_1$. So putting this together gives $$||u||_\infty \le \epsilon ||u'||_\infty + \left(1 + \frac{1}{2\epsilon}\right) ||u||_1.$$


As explained by Pacciu, $|u(x)|\leqslant\|u'\|_\infty\,|x-y|+|u(y)|$ for every $x$ and $y$. Assume first that $\varepsilon\leqslant\frac14$ and $x\geqslant2\varepsilon$ and let us integrate this inequality from $y=x-2\varepsilon$ to $y=x$. The result is $$ 2\varepsilon |u(x)|\leqslant\|u'\|_\infty\,\int_{x-2\varepsilon}^{x}|x-y|\mathrm dy+\int_{x-2\varepsilon}^{x}|u(y)|\mathrm dy\leqslant\|u'\|_\infty\,2\varepsilon^2+\|u\|_1. $$ The same inequality holds if $x\leqslant2\varepsilon$, using the integral from $y=x$ to $y=x+2\varepsilon$.

Thus, if $\varepsilon\leqslant\frac14$, $\|u\|_{\infty}\leqslant\varepsilon \|u'\|_\infty+\frac1{2\varepsilon}\|u\|_1$. In particular, $\|u\|_{\infty}\leqslant\tfrac14\|u'\|_\infty+2\|u\|_1$ hence, for every $\varepsilon\geqslant\frac14$, $\|u\|_{\infty}\leqslant\varepsilon \|u'\|_\infty+2\|u\|_1$.

Finally, for every $\varepsilon\gt0$, $$ \|u\|_{\infty}\leqslant\varepsilon\|u'\|_\infty+\max\left\{2,\frac1{2\varepsilon}\right\}\|u\|_1. $$


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.