Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Here's the framework that I am situated in and I'm having some concerns. Let $X$ be a compact metric space and $f:X\times X\times X\to \mathbb{R}$ a continuous function that can be composed as a sum of continuous functions $f_{1},f_{2}:X\times X\to \mathbb{R}$ so that $f(x,y,z)=f_{1}(x,y)+f_{2}(y,z)$.

Is it then true that: \begin{align*} \inf_{x\in X}f_{1}(x,y)+\inf_{z\in X}f_{2}(y,z)=\inf_{x,z\in X}(f_{1}(x,y)+f_{2}(y,z))=\inf_{x,z\in X}f(x,y,z)\,\, ? \end{align*} And are the functions $\inf_{x\in X}f_{1}(x,\cdot):X\to \mathbb{R}$ continuous?

share|cite|improve this question
It seems to me that $y$ plays no role at all, except for the last sentence (where $f$ plays no role at all). – André Caldas Apr 18 '12 at 11:47
You're right that $y$ does not play any role in the first question: my question is exactly whether the equality is true for all $y\in X$. And the second question concerns only $f_{1}$ and similarly $f_{2}$, not the function $f$. – T. Eskin Apr 18 '12 at 11:53
Fix $y$. Call $g_1(x) = f_1(x,y)$ and do the same for $g_2$. Then, unless I am missing some thing, you are simply asking if $\inf (g_1(x) + g_2(z)) = \inf g_1(x) + \inf g_2(z)$. You can probably easily find a counter-example for this equality. I guess the $y$ was getting in your way... :-) – André Caldas Apr 18 '12 at 13:56
up vote 1 down vote accepted

for each $x,y,z \in X$ we have \[ \inf_x f_1(x,y) + \inf_z f_2(y,z) \le f_1(x,y) + f_2(y,z) \] so the left hand side is a lower bound for $\{f(x,y,z)\mid x,z \in X\}$. As the infimum is the greatest lower bound, we obtain \[ \inf_x f_1(x,y) + \inf_z f_2(y,z) \le \inf_{x,z} \bigl(f_1(x,y) + f_2(y,z) \bigr). \] On the other hand, we have for each $x,y,z \in X$ \[ \inf_{x,z} \bigl(f_1(x,y) + f_2(y,z) \bigr) \le f_1(x,y) + f_2(y,z) \] As this holds (fixing $z$ for the moment) for each $x$ we get \[ \inf_{x,z} \bigl(f_1(x,y) + f_2(y,z) \bigr) \le\inf_x\bigl( f_1(x,y) + f_2(y,z) \bigr) = \inf_x f_1(x,y) + f_2(y,z) \] This holds for each $z$ and therefore \[ \inf_{x,z} \bigl(f_1(x,y) + f_2(y,z) \bigr) \le\inf_z\bigl( \inf_x f_1(x,y) + f_2(y,z) \bigr) = \inf_x f_1(x,y) + \inf_z f_2(y,z). \] So the inequality in question holds always, we don't need coninuity or compactness.

Regarding your second question: Since $X^2$ is compact, $f_1$ is uniformly continuous. Now let $\epsilon > 0$. By uniform continuity there is $\delta > 0$ s. t. $|f_1(x',y') - f_1(x,y)| < \epsilon$ for $d(x',x), d(y',y) < \delta$. Now let $y,y' \in X$ with $d(y',y) < \delta$. Since $X$ is compact, infima of continuous functions are attained and therefore there are $x, x' \in X$ with $g(y) = f_1(x,y)$ and $g(y') = f_1(x',y')$. Assume wlog $g(y) < g(y')$, we have \begin{align*} g(y') &= f_1(x',y')\\\ &= \inf_a f_1(a,y')\\\ &\le f_1(x,y')\\\ &\le f_1(x,y) + \epsilon \quad\text{as $d(y,y') < \delta$}\\\ &= g(y) + \epsilon\\\ \iff g(y') - g(y) &\le \epsilon \end{align*} So $g$ is (uniformly) continuous.

share|cite|improve this answer
thanks a lot martini. – T. Eskin Apr 18 '12 at 13:06

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.