3
$\begingroup$

Background: Let $x,y\in\mathbb (0,+\infty)^n$. $f$ is a continuous function on $\mathbb R^n_+=(0,+\infty)^n$. Consider the following condition (1), the sign of $f(x+y)$ is dependent on the sign of $f(x)$ and $f(y)$, formally:

$f(x)\geq0$ and $f(y)\geq\mathrel{(>)}0$ implies that $f(x+y)\geq\mathrel{(>)}0$,

$f(x)\leq0$ and $f(y)\leq\mathrel{(<)}0$ implies that $f(x+y)\leq\mathrel{(<)}0$.

Hypothesis: $f$ satisfies condition (1) if and only if there exists a linear function $g(x)=c\cdot x$ such that: $g(x)\geq0 \iff f(x)\geq 0$.

Motivation: It is known from Riesz's representation theorem that if $f(x)+f(y)=f(x+y)$, then $f(x)=c\cdot x$. Here is an effort to generalize the theorem by weakening the prerequisites.

Riesz's theorem require that the value of $f(x+y)$ is fully determined by $f(x)$ and $f(y)$. I only require that the sign of $f(x+y)$ is fully determined by $f(x)$ and $f(y)$. Although this hypothesis seems "exaggeratedly wrong" to me, I have yet to find a counterexample after thinking for a few monthes.

I also believe that such a simple idea has been thought before. Any ideas or references will help.

One might start with the facts that:

  1. If $f(x)=f(y)=0$, then $f(x+y)=f(x)+f(y)$, just like Riesz's theorem.
  2. If $f(x)>0$, then $f(\frac{x}{2})>0$.
  3. If $f(x)>0$ and $f(y)>0$, then $f(ax+by)>0$ for any $a>0$ and $b>0$.
$\endgroup$
7
  • $\begingroup$ $f(x,y)$ in first paragraph should be $f(x+y)$? $\endgroup$ Commented Dec 26, 2021 at 23:42
  • $\begingroup$ @GerryMyerson Many thanks! Fixed $\endgroup$
    – High GPA
    Commented Dec 26, 2021 at 23:43
  • 1
    $\begingroup$ What is $\mathbb R_+$? Is it $(0, \infty)$, $[0, \infty)$, or something else? $\endgroup$
    – LSpice
    Commented Dec 27, 2021 at 0:04
  • $\begingroup$ @LSpice Many thanks for noting this. Edited. $\endgroup$
    – High GPA
    Commented Dec 27, 2021 at 0:12
  • 4
    $\begingroup$ The regions where $f$ is positive or negative are convex and disjoint, so you can use the hyperplane separation theorem. $\endgroup$ Commented Dec 27, 2021 at 0:24

0

You must log in to answer this question.