Let $(U_t)_{t \in \mathbb{R}}$ be the (translation) $C_0$-group on $L^1(\mathbb{R})$ defined by $$ U_t(f)(x) = f(x-t) \quad \text{for almost every } x \in \mathbb{R} $$ (for $t \in \mathbb{R}$ and $f \in L^1(\mathbb{R})$).

The Wiener Tauberien theorem states that if $f \in L^1(\mathbb{R})$ and its Fourier transform $\hat{f}$ has no zeros on $\mathbb{R}$, then $$ cl\left( \operatorname{span}\left\{ U_t f: \, t \in \mathbb{R} \right\} \right) = L^1(\mathbb{R}), $$ where $cl(\cdot)$ is the closure of a set in the norm topology on $L^1(\mathbb{R})$.


Is there a "truncated version" of the Wiener theorem, which gives a (reasonable) sufficient condition on $f \in L^1(\mathbb{R})$ so that, for a fixed $N\in \mathbb{Z}^+$, there exists $\epsilon>0$ (depending on $N$) such that the set $$ cl\left( \left\{ \sum_{i=1}^N \beta_i U_{t_i}(f):\, t_1,\dots,t_N \in \mathbb{R},\, \beta_1,\dots,\beta_N \in \mathbb{R} \right\} \right) $$ is $\epsilon$-dense in $L^1(\mathbb{R})$.

Here, a set $Z\subseteq X$ is called $\epsilon$-dense in a Banach space $X$ if for every $x \in X$ there exists some $z \in Z$ satisfying $d(x,z)\leq \epsilon$.

The motivation for my question comes as follows:

Classical Proof of Wiener's Theorem:

In the operator-Theoretic proof of J. van Neerven, can be summarized as follows:

  • Define the Banach algebra homomorphism $L^1(\mathbb{R})\rightarrow \mathcal{L}(L^1(\mathbb{R}))$ by $$ U(f)g\triangleq \int_{-\infty}^{\infty} f(t) U(t)(g)(x) dt = f \star g, $$ (this is the convolution operator)
  • Define $X\triangleq cl(\operatorname{span}\{U_t f: \, t \in \mathbb{R}\})$ and write $U^Y$ to be the quotient operator on $L^1(\mathbb{R})/X$,
  • Show that $f \in Sp(U^Y)$ where $Sp(U^Y)$ is the Averson spectrum of $U^Y$ defined by $$ \left\{ \xi \in \mathbb{R}:\, \hat{g}(\xi)=0\, \forall g \in U^Y \mbox{ with } U^Y(g)=0 \right\}, $$
  • Use the fact that $Sp(U^Y)=0$ only if $U^Y=\{0\}$ to conclude that $Y=\{0\}$ since $f \in Sp(U^Y)$ and $\hat{f}$ was assumed to have no roots.

Edit: It's clear to me that this proof does not generalize to cover my question since the span operator cannot if $cl\left( \operatorname{span}\left\{ U_t f: \, t \in \mathbb{R} \right\} \right)$ is replaced by $cl\left\{ \sum_{i=1}^N \beta_i U_{t_i}(f):\, t_1,\dots,t_N \in \mathbb{R},\, \beta_1,\dots,\beta_N \in \mathbb{R} \right\}$? However, is there another approach to proving this theorem which can be modified to get around this added requirement?


No, you have the simple counter example: $$g_{T}(x)=\frac{1}{T}1_{0\leq x\leq T}. $$ Since $f\in L^1$, there exist $M>0$ with $\|f|_{[-M,M]}\|_{L^1}=\|f\|_{L^1}-\epsilon/N$. Moreover for any $(\beta_i,t_i)_{i\leq N}$ we have $$\|g_T - \sum \beta_i U_{t_i}f|_{[-M,M]}\|_{L^1}\geq \frac{T-2NM}{T} $$ that goes to $1$ as $T\rightarrow \infty$.


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.