Extension of Sato's hyperfunctions (Q2430133)

Fourier hyperfunctions have been introduced by T.~Kawai to provide a general and natural frame for the Fourier transformation. In fact, these Fourier techniques may also be applied to general hyperfunctions. This is due to the important result that any hyperfunction \([u]\in\mathcal B(\mathbb R) := H(\mathbb C \backslash\mathbb R)/H(\mathbb C)\) may be extended to a Fourier hyperfunction, that is, \([u]\) admits a representing function \(h\in [u]\) which is of exponential type \(0\) on each strip outside the real axis, i.e., for all \( j\in\mathbb N\) there exists \( C_j > 0 \) such that \[ \big|h(z)\big| \leq C_je^{|z|/j}\quad\text{if}\quad \frac{1}{j}\leq|\text{Im}\,z|\leq j\,. \] In fact, the following stronger result of Kaneko and Komatsu holds: Let \(F\subset\mathbb C\backslash\mathbb R\) be closed and \(k\in\mathbb N\). Then any hyperfunction \([u] \in\mathcal B(\mathbb R)\) has a representing function \(h\in [u]\) such that \(|h(z)|\leq 1/k\) on \(F\). This theorem was stated without proof by M. Sato. Classical proofs are based on Hörmander's \(\overline\partial\)-techniques (by an idea of B. A. Taylor) or on the flabbiness of the sheaf of hyperfunctions which is usually also proved by solving the \(\overline\partial\)-equation. In the present paper, the author presents a new short proof of this central result using the well-known Köthe duality and a surjectivity criterion for operators in Fréchet spaces, and then discusses the natural question: Which are the possible bounds for representing functions of general hyperfunctions? A more refined version of the above argument leads to the following characterization: Let \(F_n\), \(n\in\mathbb N\), be a closed exhaustion of \(\mathbb C \backslash\mathbb R\). Let \(\mathcal V := \{v_n\, |\, n\in\mathbb N\}\) be an increasing system of continuous (weight) functions \(v_n :\mathbb C \backslash\mathbb R\to [0,\infty[\). Define the Fréchet space \(\mathcal H\mathcal V(\mathbb C\backslash\mathbb R)\) by \[ \mathcal H\mathcal V(\mathbb C\backslash\mathbb R) :=\Big\{f\in H(\mathbb C\backslash\mathbb R)\;\Big|\; \| f\|^{}_n:= \sup_{x\in F_n} |f(z)|e^{v_n(z)} <\infty\quad\text{for all}\quad n\in\mathbb N\Big\}\,. \] Using some mild assumptions on \(F_n\) and \(\mathcal V\), the author gets the following theorem: The following statements are equivalent: (\text{a}) any \([u]\in\mathcal B(\mathbb R)\) has a representing function \(h\in\mathcal H\mathcal V(\mathbb C \backslash\mathbb R)\); (\text{b}) \(\mathcal H\mathcal V(\mathbb C \backslash\mathbb R)\cap H(\mathbb C)\neq0\); (\text{c}) \(\mathcal H\mathcal V(\mathbb C \backslash\mathbb R)\cap H(\mathbb C)\backslash\{0\}\neq0\). Though there is no representing function of Dirac's \(\delta\)-distribution decaying at in\-fi\-ni\-ty faster than \(1/|z|\), the author finally gets the following surprising improvement of the theorem of Kaneko and Komatsu: Let \(F\subset\mathbb C \backslash\mathbb R\) be closed and \(k\in\mathbb N\). Then any hyperfunction \([u]\in\mathcal B(\mathbb R)\) has a representing function \(h\in [u]\) such that \(|z^kh(z)|\leq 1/k\) on \(F\) and such that \(|z^nh(z)|\) is bounded on \(F\) for any \(n\). For the Fourier and the Laplace transformation of hyperfunctions, it is interesting to notice that one can always find a representing function in the space \[ \mathcal H_{-\infty}(\mathbb C\backslash\mathbb R):=\bigg\{f\in H(\mathbb C\backslash\mathbb R)\;\bigg|\; \sup_{1/k\leq|\text{Im\,} z|\leq k} |f(z)|e^{k|\text{Re}\,z|} <\infty\quad\text{for all}\quad k\in \mathbb N\bigg\}. \] Using the extension of \([u]\in\mathcal B(\mathbb R)\) to \(\overline{[u]}\in\mathcal H_{-\infty}(\mathbb C \backslash\mathbb R)/ \mathcal H_{-\infty}(\mathbb C)\), one can thus define an asymptotic Fourier transform on \(\mathcal B(R)\). Using the \(\pi\)-tensor product, it is no major effort to transfer the results to hyperfunctions with values in Fréchet spaces. This is interesting for applications of the Laplace transform to the abstract Cauchy problem for closed operators in Fréchet spaces. Similarly, results for hyperfunctions in several variables can be obtained from the single variable case considered so far. Both extensions of the preceding results are shortly discussed.