2
$\begingroup$

Let $k(t) \in \mathcal{L}(E; E)$, $t \geq 0$ be a family of compact operators on a Banach space $E$, such that $$ \int_0^\infty \lVert k(t)\rVert dt < \infty. $$ Let $r(t) \in \mathcal{L}(E; E)$ be the solution of integral equation $$\forall t \geq 0, \quad r(t) = k(t) + \int_0^t{ k(t-s) r(s) ds}. $$ The integral with respect to $s$ is well-defined as an Bochner integral and $k(t-s)r(s)$ is the composition of $k(t-s)$ and $r(s)$. For $\Re(z) \geq 0$, denote by $\hat{k}(z) = \int_0^\infty{e^{-zt} k(t) dt}$ the Laplace transform of $k$. Assume that $$\forall z \in \mathbb{C} \text{ with } \Re(z) \geq 0, \quad I - \hat{k}(z) \text{ is invertible}, $$ where $I$ stands for the identity operator on $\mathcal{L}(E; E)$. It is true that $\int_0^\infty{\lVert r(t)\rVert dt} < \infty$ ?


It is easy to check that the condition on the Laplace transform is required for $r \in L^1(\mathbb{R}_+; \mathcal{L}(E;E))$. Indeed, when $r \in L^1(\mathbb{R}_+; \mathcal{L}(E;E))$, we have for $\Re(z) \geq 0$: $(I+\hat{r}(z))(I-\hat{k}(z)) = (I-\hat{k}(z))(I+\hat{r}(z)) = I $, so $I-\hat{k}(z)$ is invertible. When $dim(E) < \infty$, it is known that this condition is sufficient by the Paley–Wiener theorem [1, Th. 4.1]. I am particularly interested by the case where for all $t \geq 0$, $k(t)$ is a Hilbert–Schmidt operator on a separable Hilbert space $E$. Also, I would be already be happy with the slightly weaker conclusion: $$ \forall \epsilon > 0, \quad \int_0^\infty{e^{-\epsilon t } \lVert r(t)\rVert dt } < \infty. $$

[1]: Gripenberg, G. ; Londen, S.-O. ; Staffans, O. Volterra integral and functional equations.

$\endgroup$
4
  • 1
    $\begingroup$ Spelling note: almost certainly you meant Paley–Wiener (not ‘Palay’). I edited accordingly. $\endgroup$
    – LSpice
    Commented Mar 20, 2023 at 14:48
  • 1
    $\begingroup$ If we are in a separable Hilbert space $H$, why don't we just use a family of finite-dimensional spaces $H_n$ expanding to $H$ and the corresponding family of orthogonal projections $P_n$? We have $\|P_nk(t)P_n-k(t)\|\to 0$ point-wise (due to compactness) and $\|k(t)\|$ is the integrable majorant, so $P_nkP_n$ tends to $k$ in $L^1$, after which we can approximate $P_nkP_n$ (and, thus, $k$) by a $C_0^\infty$ operator-valued function $k_1(t)$ in $L^1$ and mimic the Newman proof of Wiener's theorem (ams.org/proc/1975-048-01/S0002-9939-1975-0365002-8/… )? $\endgroup$
    – fedja
    Commented Mar 21, 2023 at 5:10
  • $\begingroup$ Thank you. I tried to do something in this spirit. If $k_n(t) = P_n k(t) P_n$, I believe that $I - \hat{k}_n(z)$ is indeed invertible for all $\Re(z) \geq 0$, provided n is large enough. So by the standard Paley-Wiener's theorem, there exists $r_n \in L^1(\mathbb{R}_+)$ such that $r_n(t)= k_n(t) + \int_0^t{k_n(t-s) r_n(s) ds}$. Now my main problem is how to check that $sup_{n \in \mathbb{N}} \int_0^\infty{ \lVert r_n(t) \rVert dt} < \infty$ ? Did I miss something? I don't see how to mimic the proof of Newman here. $\endgroup$
    – Quentin
    Commented Mar 22, 2023 at 13:42
  • $\begingroup$ "I don't see how to mimic the proof of Newman here." Just follow it literally until you hit the non-commutativity issue and show us where you get. There is a small trick in the end, which is to freeze the non-smooth entries before looking at the FT of the convolution, but let's go over Newman's scheme first: it is very elegant and powerful, so I'd rather let you try it yourself. BTW, when responding to me, you'd better add @fedja somewhere, so that I would be alerted that something is going on :-) $\endgroup$
    – fedja
    Commented Apr 1, 2023 at 2:37

0

You must log in to answer this question.

Browse other questions tagged .