Let $f: [0,1]\to \mathbb{R}$ be a lower semi-continuous function, then

$$ \liminf_{x\to a} f(x) \geq f(a), \forall a \in [0,1]$$

I have to prove that $f$ attains its minimum on $[0,1]$, that is:

$\exists x_0 \in [0,1]$ such that $f(x_0) \le f(x)$, $\forall x \in [0,1]$.

My proof:

Assume that $f$ is lower semi-continuous, but $f$ doesn't attain it's minimum on $[0,1]$,

Since $f$ is lower semicontinuous at $x_0$,

$$\forall \epsilon > 0, \exists \delta > 0 \mbox{ such that } |x - x_0| < \delta, \Rightarrow f(x) > f(x_0) - \epsilon, \forall x \in (x_0 - \delta,x_0 + \delta)$$

Now since $f$ doesn't attain it's minimum on $[0,1]$, then

$$\forall x_0 \in [0,1], \exists x \in [0,1] \mbox{ s.t } f(x_0) > f(x)$$

Let $\epsilon = f(x_0) - f(x) > 0, \exists \delta > 0$, such that $|x-x_0| < \delta$ and $f(x) > f(x_0) - \epsilon \Rightarrow f(x_0) - f(x) > \epsilon = f(x_0) - f(x)$ which is a contradiction.

I am right? Thanks.

  • 1
    $\begingroup$ It's wrong. You should have said "$\forall\epsilon>0\forall x_0\in [0,1] $... such that $|x-x_0|<\delta \implies f(x)>f(x_0)\color{red}- \epsilon$". $\endgroup$ Jan 21 '19 at 17:04
  • 1
    $\begingroup$ It is $-\epsilon$ in the definition, and the $\forall x$ has to be after the “such that”. $\endgroup$
    – Mindlack
    Jan 21 '19 at 17:04
  • $\begingroup$ Yes, I fixed the problem about the epsilon, it was a typo using the upper semi-continuity. I will see the quantifier problem now. $\endgroup$ Jan 21 '19 at 17:06
  • 1
    $\begingroup$ No. The $\delta$ is allowed to depend on $\epsilon$ and $x_0$, not $x$. $\endgroup$
    – Mindlack
    Jan 21 '19 at 17:16
  • 2
    $\begingroup$ An idea to fix this proof: if $f$ does not attain a minimum, there is a sequence $x_n$ in $[0, 1]$ such that $f(x_n) > f(x_{n+1})$ for every $n$. Now, this sequence must have a convergent subsequence since $[0, 1]$ is compact. Try to reach your contradiction from here. $\endgroup$
    – Daniel
    Jan 21 '19 at 17:34

Since $f$ is lower-semicontinuous, for each $x\in [0,1],$ there is an open interval $I_x\subseteq [0,1]$ such that $\inf\{f(y):y\in I_x\}\ge f(x)-1.$ The $I_x$ form an open cover of $[0,1]$ so passing to a finite subcover, we conclude that $f$ is bounded below.

So, letting $y=\inf\{f(x):x\in [0,1]\}$, we can find a sequence $(x_n)\subseteq [0,1]$ such that $f(x_n)\to y.$ And of course, there is a subsequence $(x_{n_k})\subseteq (x_n)$ such that $x_{n_k}\to x_0\in [0,1]$.

Then, $f(x_0)\le \liminf_{x\to x_0} f(x)\le \liminf_{k\to \infty}f(x_{n_k})=\lim_{k\to \infty}f(x_{n_k})=y,$ which implies that $f(x_0)=y.$


Another characterization of lower semi-continuous function that might be helpful:

A function $f:[0,1] \to \Bbb R$ is lower semi-continuous if and only if its upper level set $\{x\in [0,1]:f(x)>c\}$ is open in $[0,1]$ for all $c\in\Bbb R$.

(Indeed above characterization can be applied to every topological space $X$ and $f:X\to\Bbb R$. And the proof below can be generalized to any compact topological space.)

Proof. Assume $x_0\in X$ is such that $f(x_0)>c$. If we choose $0<\epsilon< f(x_0)-c$, we can find $\delta>0$ such that $$x\in [0,1], x\in (x_0-\delta,x_0+\delta)\implies f(x)>f(x_0)-\epsilon.$$ Since $f(x_0)-\epsilon>c$, it follows $$ [0,1]\cap (x_0-\delta,x_0+\delta)\subset \{x : f(x)>c\}. $$ This proves $\{x:f(x)>c\}$ is open in $[0,1]$. $\blacksquare$

By the above proposition, $F_c=\{x:f(x)\le c\}$ is closed in $[0,1]$ for every $c\in\Bbb R$, hence is compact. If there is no minimum, then $F_c$ is not empty for every $c$. Consider the decreasing sequence of closed sets$$F_{-1}\supset F_{-2}\supset \cdots\supset F_{-n}\supset\cdots.$$ Since $$ \bigcap_{n\in\Bbb N}F_{-n} = \varnothing, $$ by the finite intersection property of the compact set $[0,1]$, there exists $n_0\in\Bbb N$ such that $F_{-n_0}=\varnothing$. This implies $f(x)\ge -n_0$ for all $x\in [0,1]$, hence $f$ is lower bounded.

Let us define $S=\{c\in\Bbb R: F_c\ne \varnothing\}$. Since $-n_0\notin S$ and $-n_0\le S$, we know $S$ is lower bounded. Let $m=\inf S$. Then by finite intersection property, $$ F_m=\bigcap_{k\in\Bbb N}F_{m+1/k}\ne \varnothing. $$ Thus there exists $x_0$ such that $f(x_0)=m$. Finally, for all $c<m$, it holds $c\notin S$, i.e. $F_c=\varnothing.$ This establishes $m$ is the minimum of $f$.

  • $\begingroup$ In this proof, did we use that $F_c$ is compact? Thanks! $\endgroup$ Dec 5 '20 at 7:25

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.