Let $I=[0,1]$ and $E$ a Banach space. We note by $X:=\mathcal {C}(I,E), $ the space of all continuous functions from $I$ to $E$, with $\left \| x \right \|_X=\sup_{t\in I }\left \| x(t) \right \|_E $.

Let $f:I\times E\rightarrow E$ a function such that:

  • For each continuous $x\in X$, we have $f(.,x(.))$ is Pettis integrable on $I$,

  • for every $t \in I,\:\: f_t: E \rightarrow E,\:u \mapsto f_t(u):=f(t,u) \text{ is continuous}.$

Let $$T: X \rightarrow X,\:x \mapsto T(x)(t):=\int_{0}^{t}f(s,x(s)) ds$$

Claim: $T$ is continuous.

This is how I tried to solve this:

For $t\in I,\:f_t$ is continuous, that is,

for each $u\in E$, $\forall \epsilon>0 , \exists \eta_{t,u,\epsilon}>0 \text{ such that } \forall v\in E$ $$\left \|u-v \right \| \leq \eta_{t,u,\epsilon} \Rightarrow \left \| f(t,u)-f(t,v) \right \| < \epsilon $$

Now, let $t\in I$, $\epsilon >0$ , and $x\in X$. Let $y\in X$ such that $$\left \| x-y \right \|_X\leq \eta_{t,x(t),\epsilon}\;,$$

i.e. $$\forall s\in I,\:\left \| x(s)-y(s) \right \|_{E\times E}\leq \eta_{t,x(t),\epsilon}\;,$$ in particular, $$\left \| x(t)-y(t) \right \|_{E\times E}\leq \eta_{t,x(t),\epsilon}\;.$$

Hence, $$\left \| f(t,x(t))-f(t,y(t)) \right \| < \epsilon \quad(*) $$

So, $$\begin{matrix} \left \| T(x)(t)-T(y)(t) \right \| & = &\left \| \int_{0}^{t} f(s,x(s))-f(s,y(s)) ds\right \| \\ & \leq & \int_{0}^{t} \left \| f(s,x(s))-f(s,y(s)) ds\right \| \quad(**)\\ \end{matrix}$$

unfortunately, I can't use $(*)$ in $(**)$ because it $(*)$ not uniformaly on $t$.

Is our claim true? why?

If not, what is the condition on $f_t$ that you suggest instead of continuity?


I do not have a counterexample, but a strong feeling that the conjecture is false, based on the following positive proof.

If you require slightly more, namely Lebesgue integrability of $t\mapsto f(t,x(t))$ for every $x\in Z=L_\infty([0,1],E)$, then it is well-known that the superposition operator $F(x)(t)=f(t,x(t))$ is “automatically” continuous from $Z$ into $Y=L_1([0,1],E)$ (because $f$ is a Carathéodory function, and $Y$ is a regular ideal space), see e.g. Theorem 5.2.1 in M. Väth, Ideal spaces, Springer, 1997.

This implies the continuity of $T\colon Z\to X$, because if $x_n\to x$ in $X$ then $x_n\to x$ in $Z$, and so

$$\lVert T(x_n)-T(x)\rVert_X\le\lVert F(x_n)-F(x)\rVert_Y\to0\text.$$

I am not sure whether the automatic continuity of $F$ holds also in case $Z=C([0,1],E)$, but I recall that such a result was cited in case $E=\mathbb R$, and I was unable to access it. However, probably the above mentioned proof can be modified straightforwardly for that case.

Anyway, the proof in the “simpler” case $Z=L_\infty([0,1],E)$ already depends on a subtle application of Vitali's convergence theorem (much more subtle than the straightforward attempt in the question), and it is hard to imagine that there is any proof which comes without using some sort of equi-integrability of the sequence $F(x_n)$. Since the latter is exactly what you lack for Pettis integrability, I would be very surprised if the assertion would be true.

| cite | improve this answer | |
  • $\begingroup$ Thank you for your answer, and for the reference. I hhave some remarks here: 1) Since $\left \| T \right \|_Z\leq \left \| . \right \|_X$, why $\left \| T(x_n)-T(x) \right \|_X\leq \left \| F(x_n)-F(x) \right \|_Z $ ? 2)Do you suggest that I suppose that $t \mapsto f(t,x(t)) $ is Pettis integrable for every $x\in Z$ instead of $x\in X$? $\endgroup$ – Motaka 2 days ago
  • $\begingroup$ 1. For every $t\in[0,1]$ there holds $\lVert T(x_n)(t)-T(x)\rVert_E\le\int_0^t\lVert F(x_n)(s)-F(x)(s)\rVert_E\,ds\le\lVert F(x_n)-F(x)\rVert_{L_1([0,1],E)}$, hence the right-hand side is a bound for the maximum over all $t\in[0,1]$. $\endgroup$ – Martin Väth 2 days ago
  • $\begingroup$ 2. No, I assume that for the Pettis integral there is no result at all about automatic continuity, even under the stronger hypothesis you mention. $\endgroup$ – Martin Väth 2 days ago
  • $\begingroup$ A misunderstanding made me think that $F$ and $T$ are the same up to the starting space. For $2$, unfortunately, the conditions that I have are those mentioned in the question. $\endgroup$ – Motaka 2 days ago

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.