Show that $\sup (A\cdot B)=\max\{\sup A\cdot\sup B, \sup A\cdot\inf B,\inf A\cdot\sup B,\inf A\cdot\inf B\}$

Given nonempty subsets $A$ and $B$ of positive real numbers, define $$A\cdot B=\{z=x\cdot y:x\in A,\,y\in B \}$$ show that if $A$ and $B$ are bounded sets of real numbers, then $$\sup(A\cdot B)=\max\{\sup A\cdot\sup B, \sup A\cdot\inf B,\inf A\cdot\sup B,\inf A\cdot\inf B\}$$

For this question, I know that $\sup(A\cdot B)=\sup(A)\cdot\sup(B)$. Do I need to compare all elements in the set then show $\sup(A\cdot B)=\sup(A)\cdot\sup(B)$? Am I on the right way? If not, can anyone give a hit or a suggestion to start? Thanks.

• It’s not necessarily true that $\sup(A\cdot B)=\sup A\cdot\sup B$: consider $A=\{-2,-1\}$, $B=\{-1\}$. You’ll need to take into account algebraic signs. – Brian M. Scott Jul 4 '15 at 22:24
• Can't you brute force this by considering cases? $A,B>0$ etc. – Matematleta Jul 4 '15 at 23:17
• @BrianM.Scott The sets A and B given are of positive real numbers. – marchimedes Oct 25 '16 at 19:34
• @marchimedes: Not in the actual question: it’s asks simply about bounded sets of real numbers. Moreover, there would be no need to take the maximum of those four products if $A$ and $B$ were required to be sets of positive real numbers. I suspect that positive in the first definition isn’t supposed to be there. – Brian M. Scott Oct 25 '16 at 19:40

For any $x\in A, y\in B$, there is $$\inf{A}\leqslant x\leqslant \sup{A} \hspace{4 mm} \text{and} \hspace{4 mm} \inf{B}\leqslant y\leqslant \sup{B}$$ So $$xy\leqslant\max\{\sup A\cdot\sup B, \sup A\cdot\inf B,\inf A\cdot\sup B,\inf A\cdot\inf B\}$$ And $$xy\geqslant\min\{\sup A\cdot\sup B, \sup A\cdot\inf B,\inf A\cdot\sup B,\inf A\cdot\inf B\}$$ Thus $$\sup{A\cdot B}\leqslant\max\{\sup A\cdot\sup B, \sup A\cdot\inf B,\inf A\cdot\sup B,\inf A\cdot\inf B\}$$ And $$\inf{A\cdot B}\geqslant\min\{\sup A\cdot\sup B, \sup A\cdot\inf B,\inf A\cdot\sup B,\inf A\cdot\inf B\}$$

• It is not true that $\sup(A\cdot B)=\sup A\cdot \sup B$ if $\sup A>0$ and $\sup B>0$. See for example $A=\{1,-2\},B=\{1,-2\}$ then $\sup(A\cdot B)=4$. – Listing Jul 5 '15 at 1:17
• @Listing is correct. That equation only holds for non-negative sets of real numbers (of course, non-empty and bounded above). – marchimedes Oct 25 '16 at 19:53

Let the inf and sup of $A$ be $a_1,a_2$ and of $B$ be $b_1,b_2.$ Now let $R$ denote the rectangle $R=[a_1,a_2]\times [b_1,b_2],$ and we seek the sup of $p=xy$ over $R.$

We first assume that $A,B$ are the entire closed intervals between their respective inf and sup, and say later why this is OK. Note first that the only critical point of $p$ is at the origin. If this critical point happens to lie interior to $R$ then each of $a_2,b_2$ are positive, and since they are sups, there are sequences approaching them, so that the sup we seek is at least $a_2b_2>0,$ which means the sup of $p$ is positive, so the maximum of $p$ on $R$ does not occur at the critical point of $p$.

So the max of $p$ must occur at a boundary point of $R,$ and focusing on any of the four edges of $R$ then shows $p$ cannot have its max at an interior point of an edge of $R.$ This leaves only the four vertices of $R$ as candidates for the max of $p$ over $R.$ So by the usual method of maximizing a two variable differentiable function over a compact domain, the maximum of $p$ over $R$ is the greatest product among $a_1b_1,\ a_1b_2,\ a_2b_1,\ a_2b_2$ which in our notation is the desired supremum of $p.$ Call this amount $M.$

Slight correction: If an edge of $R$ happens to be along the $x$ or $y$ axis, then $p$is constant ($p=0$) on that edge. So above, instead of saying the max of $p$ cannot occur at an interior point of an edge, I should have said more correctly something like "The max of $p$ is taken on at an endpoint of the edge (and perhaps at interior points, if the edge is along an axis)."

Now of course the given sets $A,B$ need not be intervals, but by the above argument the product $p$ taken at any point $(x,y) \in A \times B$ will be at most $M$. We can be sure the supremum is actually equal to $M$ since whichever "corner" of the rectangle is chosen, there are sequences in $A,B$ approaching the coordinates of that corner.