Let $f:\mathbb R^n\to \mathbb R\cup\{+\infty,-\infty\}$. The limit inferior of $f$ at $x_0 \in \mathbb R^n$ is defined as \begin{equation} \liminf_{x\to x_0}f(x) = \lim_{\varepsilon \downarrow 0}\left(\inf\{f(x)\mid x \in (x_0 + \varepsilon B_n)\setminus\{x_0\}\}\right) \end{equation} where $B_n$ is the closed unit ball. I am trying to prove or disprove the following conjecture:

There exists a sequence $\{x_k\}$ of $\mathbb R^n$ such that $x_k \to x_0$ and $f(x_k) \to \liminf_{x\to x_0} f(x)$ as $k \to \infty$.

My first trial was as follows:

Define $U_k = (x_0 + \varepsilon_k B_n)\setminus\{x_0\}$ and $m_k = \inf\{f(x)\mid x \in U_k\}$ with $\varepsilon_1 > \varepsilon_2 > \cdots$ and $\varepsilon_k \to 0$. Then, $m_1 \le m_2 \le \cdots$ and $m_k \to \liminf_{x\to x_0}f(x)$. If we pick a point $x_k$ from a set $V_k = \{x\in U_k\mid m_k \le f(x) \le m_{k+1}\}$, then $x_k \to x_0$ and $f(x_k) \to \liminf_{x\to x_0}f(x)$.

However, if $\liminf_{x\to x_0}f(x) = \inf\{f(x)\mid x \in U_1\}$ and $\liminf_{x\to x_0}f(x) \notin f(U_1)$, then $V_k = \emptyset$ and the proof does not work. Would you give me any hint or reference?

=================================================================== Edit:

I have corrected the proof thanks to Martin Argerami's answer.

Define $U(\varepsilon) = (x_0 + \varepsilon B_n)\setminus\{x_0\}$ and $m(\varepsilon) = \inf\{f(x)\mid x \in U(\varepsilon)\}$. Then, $m(\varepsilon) \to m_0 = \liminf_{x \to x_0}f(x)$ as $\varepsilon \to 0$.

1) If there exists $\varepsilon > 0$ such that $m(\varepsilon) = +\infty$, then $f(x) = +\infty$ for every $x \in U(\varepsilon)$, so $f(x_k) \to m_0 = +\infty$ for every sequence $\{x_k\}$ that converges to $x_0$.

Suppose $m(\varepsilon) < +\infty$ for all $\varepsilon >0$.

2) If there exists $\varepsilon_1 >0$ such that $m(\varepsilon) = -\infty$ for every $0 < \varepsilon \le \varepsilon_1$, then let $\varepsilon_1 > \varepsilon_2 > \cdots$ such that $\varepsilon_k \to 0$.

2-1) If $-\infty \in f(U(\varepsilon_k))$, let $x_k \in U(\varepsilon_k)$ such that $f(x_k) = -\infty$.

2-2) If $-\infty \notin f(U(\varepsilon_k))$, let $x_k \in U(\varepsilon_k)$ such that $f(x_k) < -k$; such $x_k$ exists by the definition of the infimum.

Then $x_k \to x_0$ and $f(x_k) \to m_0 = -\infty$ as $k \to \infty$ because $f(x_k) < -k$.

3) If there exists $\varepsilon_1 >0$ such that $-\infty < m(\varepsilon_1) < +\infty$, then $-\infty < m(\varepsilon) < +\infty$ for every $0 <\varepsilon \le \varepsilon_1$. Let $\varepsilon_1 > \varepsilon_2 > \cdots$ such that $\varepsilon_k \to 0$. By the definition of the infimum, there exists $x_k \in U(\varepsilon_k)$ with $|f(x_k) - m(\varepsilon_k)| < 1/k$. Then $x_k \to x_0$ and \begin{equation} \lim_{k\to\infty}f(x_k) = \lim_{k\to\infty}m(\varepsilon_k) = m_0. \end{equation}

  • $\begingroup$ In the $-\infty$ argument, you cannot say that $f(x_k)=-\infty$; you can say that $f(x_k)$ is very negative, but it doesn't have to be infinite. $\endgroup$ – Martin Argerami Aug 31 '16 at 15:20
  • $\begingroup$ @Martin Argerami Would you explain a little bit more why $f(x_k) = -\infty$ is not possible when $f:\mathbb{R}^n\to\mathbb{R}\cup\{-\infty,+\infty\}$? I guess that if both $+\infty$ and $-\infty$ are in the range of $f$, then $\infty - \infty$ could happen but I am not sure if it is the reason. $\endgroup$ – flyingwith Aug 31 '16 at 16:12
  • $\begingroup$ It is not that it is not possible; it is not certain. For instance consider $f(t)=1/t$ and $f(0)=0$. With the definition you gave, $\liminf_{t\to0} f(t)=-\infty$, but there is no $t$ with $f(t)=-\infty$. $\endgroup$ – Martin Argerami Sep 1 '16 at 5:19
  • $\begingroup$ @MartinArgerami It will be handled by the case 2-2). $\endgroup$ – flyingwith Sep 1 '16 at 7:08

By definition of inf, for each $k $ there exists $x_k\in U_k $ with $|m_k-f (x_k)|<1/k $. Then $$ \liminf_{x\to x_0}f (x)=\lim_{k\to\infty}m_k=\lim_{k\to\infty}f (x_k). $$

You need to deal with the $\pm\infty $ case on its own, but the spirit is the same.

  • $\begingroup$ Great! Thanks a lot. $\endgroup$ – flyingwith Aug 31 '16 at 11:51
  • $\begingroup$ I have corrected the proof by your answer. $\endgroup$ – flyingwith Aug 31 '16 at 13:31

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.