I posted the following question on MSE originally because, not being research-level, it seemed more appropriate for that site. However, there was no activity for it on MSE and I feel that it certainly can be answered on MO, so I am now posting it here.

A number of conditions that are equivalent to the statement that a Banach space $X$ has Radon-Nikodym property (RNP) are found on page 217 of ``Vector Measures" by Diestel & Uhl. One such condition is that: a Banach space $X$ has the RNP iff every function $f:[0,1]\to X$ that is of bounded variation is weakly differentiable off a fixed set of measure zero.

To me, the condition that $f:[0,1]\to X$ is ``weakly differentiable off a fixed set of measure zero" means that for a.e. $t\in(0,1)$, there exists a vector $x_{t}\in X$ such that $\lim_{h\to 0}\frac{(\varphi\circ f)(t+h)-(\varphi\circ f)(t)}{h}=\varphi(x_{t})$ for every $\varphi\in X^{*}$. Note that that this interpretation of weak differentiability at a point $t\in(0,1)$ matches the definition found in this paper at the bottom of page 2802 (the second page of the introduction).

First question: Is this interpretation of the condition that $f:[0,1]\to X$ is ``weakly differentiable off a fixed set of measure zero" correct? If not, what is the correct interpretation?

For example, one might reasonably also say that $f:[0,1]\to X$ is weakly differentiable if there exists a function $g:[0,1]\to X$ such that, for every smooth test function $\varrho:[0,1]\to\mathbb{R}$ with compact $\text{supp}(\varrho)\subset(0,1)$, it follows that $\int_{0}^{1}\varrho'(t)f(t)dt=-\int_{0}^{1}\varrho(t)g(t)dt$. This is, in turn, equivalent (see here, Prop. 6.36) to the dual condition that $\int_{0}^{1}\varrho'(t)(\varphi\circ f)(t)dt=-\int_{0}^{1}\varrho(t)(\varphi\circ g)(t)dt$ for every $\varphi\in X^{*}$. Clearly, if $g:[0,1]\to X$ is defined for a.e. $t\in[0,1]$ by $g(t)=x_{t}$ (where $x_{t}\in X$ is as stated above the first question) then the dual condition holds at least formally.

Second question: Is the dual condition for weak differentiability equivalent to the interpretation of weak differentiability above the first question?


1 Answer 1


Diestel & Uhl can only mean the first interpretation you gave for two reasons:

  1. For the second interpretation, the term “off a fixed set of measure zero” makes no sense.
  2. Even in case $X=\mathbb R$ (which certainly has the RNP) the condition is not satisfied if interpreted in the second way: Cantor's stairway function $f$ is monotone and thus of bounded variation, and for $f$ the only candidate for $g$ in the second interpretation is the $0$-function.

The latter shows also that the answer to your second question is negative.

  • $\begingroup$ Thanks for your answer. I had the same thought about the phrase ``off a fixed set of measure zero" and Cantor's stairway function is a great example. $\endgroup$
    – JWP_HTX
    Apr 27, 2022 at 14:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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