3
$\begingroup$

Is this possible?

$\{f_n(t)\},n\geq1, t\in[0,1],$ is a sequence of absolutely continuous functions with $f_n(0)=f_n(1)=0.$

$$\int_0^1f_n'(t)^2dt\leq C<\infty,$$ but $$\lim_{n\rightarrow\infty}\int_0^1\left(\frac{f_n(t)}{1-t}\right)^2dt=\infty.$$

I strongly believe that this is IMPOSSIBLE, can someone help? Thanks a lot...

$\endgroup$
  • $\begingroup$ A quick fact: $\|f_n\|^2\leq C$ under the uniform norm. $\endgroup$ – user108871 Sep 17 '14 at 15:03
3
$\begingroup$

Let $0<h<1$. We have that:

$$\int_0^h \left(\frac{f_n(t)}{1-t}\right)^2dt= \left[\frac{f_n(t)^2}{1-t}\right]_0^h -2\int_0^h \frac{f_n(t)f_n^{\prime}(t)}{1-t}dt$$ We have: $$ \left[\frac{f_n(t)^2}{1-t} \right]_0^h=\frac{f_n(h)^2}{1-h}=f_n(h)\frac{f_n(h)}{1-h}$$ But $\displaystyle \frac{f_n(h)}{1-h} \to -f_n^{\prime}(1)$ as $h\to 1$, and $f_n(h)\to f_n(1)=0$. Hence we show if $h\to 1$ that $$ \int_0^1 \left(\frac{f_n(t)}{1-t}\right)^2dt= -2\int_0^1 \frac{f_n(t)f_n^{\prime}(t)}{1-t}dt$$

Now we use the Cauchy-Schwarz inequality:

$$\left(\int_0^1 \frac{f_n(t)f_n^{\prime}(t)}{1-t}dt\right)^2\leq \int_0^1 \left(\frac{f_n(t)}{1-t}\right)^2dt\int_0^1 \left(f_n^{\prime}(t)\right)^2dt$$

Hence $\displaystyle \int_0^1 \left(\frac{f_n(t)}{(1-t)}\right)^2 dt=0$ or $$\int_0^1 \left(\frac{f_n(t)}{(1-t)} \right)^2dt\leq 4\int_0^1(f_n^{\prime}(t))^2dt\leq 4c$$ and we are done.

$\endgroup$
  • $\begingroup$ Way is named Cauchy-Schwartz??? These mathematicians are from different centuries, lol. $\endgroup$ – Masacroso Sep 17 '14 at 17:05
  • $\begingroup$ does holds $\lim_{t\to 1}\frac{ f_n(t)^2}{(1-t)}=0$? to me it is not clear, since $\lim_{t\to 1^-} 1/(1-t)=+\infty$ and $f_n(0)=0$. Your solution does not considerate this. $\endgroup$ – math_man Sep 17 '14 at 17:11
  • $\begingroup$ OK, It is true ; I have edited my answer. $\endgroup$ – Kelenner Sep 17 '14 at 17:30
  • $\begingroup$ Thanks...the solution looks good! I may prefer to show $\lim_{t\rightarrow1}\frac{f_n(t)^2}{1-t}=0.$ This is true since $f_n(t)^2=(\int_t^1f_n'(s)ds)^2\leq (1-t)\int_t^1f_n'(s)^2ds.$ This convergence doesn't have to be uniform in $n.$ The thing is that $\frac{f_n(h)}{1-h}\rightarrow-f_n'(1)$ may not be well-defined at a specific point. $\endgroup$ – user108871 Sep 17 '14 at 17:51
  • $\begingroup$ @math_man: Thank you for the edition. $\endgroup$ – Kelenner Sep 17 '14 at 18:13

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.