5

I am trying to proof relative compactness in L2(0,1) for a specific set of functions $(\phi_n)_{n \in \mathbb{N}}$ with following properties:

  1. $\int_0^1 \phi(x) dx = 0 $
  2. $||\phi_n^2||_{L1(0,1)} = 1 $
  3. $(\phi_n)_{n \in \mathbb{N}}$ is uniformly integrable such that for each $\epsilon >0$ there exist a $\delta> 0$ fulfilling: $$\sup_{n \in \mathbb{N}} \int_0^1 \phi_n(x)^2*\mathbb{1}\left(|\phi_n(x)| > \delta\right) dx < \epsilon $$ Here, $\mathbb{1}$ is the indicator function resulting in 1 if the inner statement is true.
  4. Each $\phi$ is non decreasing and has at most countable many discontinuity points. (Otherwise it is continuous)

I tried to use the theorem of Kolmogorow-Riesz, where I have to proof boundedness in the norm (which is already given above) and equicontinuity: $$ \lim_{h \rightarrow 0} \sup_{n \in \mathbb{N}} \int_0^1 \left(\phi_n(x+h) - \phi_n(x)\right)^2 dx$$

But I have difficulties to show the equicontinuity, since I do not find a possibility to use the given uniform integrability. I plotted many examples and I think that this property is necessary in this scenario. Without this property I think there could be a $\phi_n$ that has all its mass at the border of 0 and 1 so that the equicontinuity can not hold anymore. I Imagine that the uniform integrability should lead to distributing the mass uniformly so that the actual difference of $\phi_n(x+h)-\phi_n(x)$ is bounded in terms of h.

Does anyone have a source where something like this is done or is there a more elegant way than Kolmogorow-Riesz for showing the relative compactness?

  • Sorry. I just edited 1) accordingly. I got rid of the absolute value of phi. – Alphacache Dec 30 '23 at 10:42
  • In 3., what is $d$? I guess that $\phi$ should be $\phi_n$ – Davide Giraudo Dec 30 '23 at 13:01
  • Doesn’t your condition 3. Imply that the sequence $\phi_n$ converges to zero in norm? – Paul Gullesh Dec 30 '23 at 15:37
  • You are right. The division by $\sqrt{n}$ is not right in my case. I corrected the question above. – Alphacache Dec 30 '23 at 16:07
  • An idea: uniform integrability in this context should probably yield uniform absolute continuity, i.e. for every $\epsilon>0$ there is $\delta>0$ such that $m(A)<\delta\implies\int_A |\phi_n|^2 < \epsilon$, which might help you prove equicontinuity of translations (for instance, this may allow you to prove that the $\phi_n$ can be uniformly approximated in $L^2$ by continuous functions). Just some thoughts, not sure if they will work. – kieransquared Dec 30 '23 at 16:42