2
$\begingroup$

I admittedly have a fairly stupid question about a problem, which I wanted to make sure I understand correctly, since I actually never came across this before. Let $f$ be a function, $f: \mathbb{R^2} \rightarrow \mathbb{R}$. What does $$\inf_{x \in \mathbb{R}} {\sup_{y\in \mathbb{R}}}f(x,y)$$ exactly mean? If i want to evaluate the supremum first, do I look at every $x$ but treat it as a constant value or do I just pick one x and evaluate $f(x,y_1),...,f(x,y_n),...$ and then take the smallest upper bound of that? And what does the infimum over that then look like? I am kind of confused about that notation, appreciate any help. Thank you in advance!

Edit: My original problem is the following. Let $g$ be an estimator for $\theta$. The risk is given by $R(\theta,g)$. I am now supposed to evaluate $\inf_{g} \sup_{\theta \in \Theta}R(\theta,g)$, where $\Theta$ is the parameter set and $g$ is an estimator. This is used in Stochastics and Probability theory.

$\endgroup$
0
1
$\begingroup$

You can think of it in the following way. First, define the function $g\colon \mathbb R \to \mathbb R$ by $$ g(x) = \sup_{y \in \mathbb R} f(x,y). $$ Then, letting $C = \inf_{x\in\mathbb R} \sup_{y\in\mathbb R} f(x,y)$, we have that $$ C = \inf_{x\in\mathbb R} g(x). $$

In other words, you can find $C$ by first finding the supremum over $y$ of $f(x,y)$ for each $x$. This gives you the function $g$ described above. Then, you can find the infimum of $g$ over $x$.

Note that, in general, $$ \sup_{y\in\mathbb R} \inf_{x\in\mathbb R} f(x,y) \le C. $$ This means you should not do the process I described above in reverse, as you may sometimes have strict inequality above. (See Wikipedia.) The conditions where you have equality is the subject of a number of different theorems. See, for example, Sion's Theorem.

$\endgroup$
0

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.