In Nowak (1989), a modular $\rho$ on a vector lattice is defined by the following properties

(N1) $\rho(x)=0\implies x=0$;

(N2) $\lvert x\rvert \le \lvert y\rvert\implies \rho(x) \le \rho(y)$;

(N3) $\rho(x\vee y )\le \rho(x)+\rho(y)$ for all $x\ge 0, y\ge 0$;

(N4) $\rho(\lambda x)\to 0$ if $\lambda \to 0$

This supposedly implies $$ \rho(\alpha x+\beta y)\le \rho( x)+\rho(y) \text{ for all } \alpha,\beta\ge0,\alpha+\beta=1,$$ which features in the definition of modular in Musielak and Orlicz (1959).

I don't see how this follows. NB: convexity is not assumed.

  • $\begingroup$ Perhaps you need to prove $$\alpha x+\beta y\le x+y \text{ for all } \alpha,\beta\ge0,\alpha+\beta=1,$$ maybe just for positive $x,y$. Is that true in a vector lattice? $\endgroup$ – Gerald Edgar Sep 5 '17 at 12:24
  • $\begingroup$ @GeraldEdgar For positive $x$ and $y$ it is certainly true. But what then? $\endgroup$ – encore Sep 5 '17 at 12:55
  • $\begingroup$ Ok, I can see how this will work for positive and disjoint elements. If $\lvert x\rvert \wedge \lvert y\rvert = 0$ then $\lvert x\rvert \vee \lvert y\rvert = \lvert x\rvert + \lvert y\rvert$ and the result follows from (N3) because $\rho(\alpha x)$ is known to be non-decreasing function of $\alpha$ on $[0,\infty)$ for any fixed $x$. $\endgroup$ – encore Sep 5 '17 at 13:09

How about this?
Let $\alpha, \beta \ge 0$, $\alpha+\beta = 1$.
Then $$ |\alpha x + \beta y| \le \alpha |x| + \beta |y| \le |x| \vee |y| $$ so $$ \rho(\alpha x + \beta y) = \rho\big(|\alpha x + \beta y|\big) \le \rho\big(|x| \vee |y|\big) \le \rho\big(|x|\big) +\rho\big(|y|\big) = \rho(x)+\rho(y) $$

Note: $\rho(x) = \rho\big(|x|\big)$ from N2.

| cite | improve this answer | |
  • $\begingroup$ You must be pretty close :-). In Musielak-Orlicz definition there is symmetry: $\rho(x)=\rho(-x)$. However, I do not think symmetry implies equality with $\rho(\lvert x \rvert)$ unless the modular is so-called orthogonally additive, i.e. unless one has $\rho(x+y)=\rho(x)+\rho(y)$ when $x,y$ are disjoint. $\endgroup$ – encore Sep 5 '17 at 13:37
  • $\begingroup$ I've had a mistake in (N2) and (N3), now they are fixed. From (N2) $\rho(x)=\rho(\lvert x\rvert)$ follows. But I think your first line cannot hold as stated, just take $x=y=1/2$ to falsify it. $\endgroup$ – encore Sep 5 '17 at 15:46
  • $\begingroup$ Oh bum, you are right, of course. Thank you! $\endgroup$ – encore Sep 5 '17 at 17:34

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.