Convolution transform for Boehmians (Q1880845)

In this paper, the authors present a theory to develop the convolution transform in the context of Boehmian spaces. Starting with a permissible convolution kernel \(G\), the authors define a Boehmian space in which the convolution transform with respect to the kernel \(G\) can be defined and studied. There are many errors, factual and typographical, in this paper. The top space for their Boehmian space is the countable union space \[ T= \bigcup^\infty_{n=1}{\mathcal L}_{-a_n, -b_n}, \] where \(\{a_n\}\) is an increasing sequence of positive real numbers converging to \(\alpha_2\) and \(\{b_n\}\) is a decreasing sequence of negative real numbers converging to \(\alpha_1\), where \(\alpha_1\), \(\alpha_2\) are defined by \(G\) and where \({\mathcal L}_{-a_n, -b_n}\) are the classical Zemanian spaces. This space \(T\) is nothing but the Zemanian space \({\mathcal L}(-\alpha_2, -\alpha_1)\) and is independent of the choice of the sequences converging to \(\alpha_2\) and \(\alpha_1\). The authors use a pair of specific sequences in their description and this leads to many avoidable mistakes throughout their paper. There is also a confusion with regard to the kernel \(G\) which gives rise to the space \(T\). The authors use frequently the expression ``a convolution kernel'' after fixing \(\alpha_1\) and \(\alpha_2\) in Theorems 4.5, 4.9, 4.12, Corollary 4.13 and Definition 4.14. In the preliminaries in Section 2, the authors claim that \[ \sup_{\substack{ x\in\mathbb{R}\\ |x|\leq m}} {k_{c,d}(x- t)\over k_{c,d}(x)}\leq e^{\max\{|c|,|d|\}m} \] without any relationship between \(c\) and \(d\). This is incorrect. One needs \(c> d\) to obtain this. Definition 2.5 is misleading. It is assumed that \(f\in{\mathcal L}_{c,d}'\) and \(\psi\in{\mathcal L}_{c,d}\) and the convolution is defined as a \(C^\infty\) function by \[ (f* \psi)(x)=\langle f,\tau_x\check\psi\rangle,\quad\text{for all }x\in \mathbb{R}. \] This definition does not make sense because, if \(\psi\in{\mathcal L}_{c,d}\), \(\tau_x\check\psi\) need not belong to \({\mathcal L}_{c,d}\). This definition is invoked in Theorem 2.8, Theorem 3.1 and in Section 4. The best way to change this definition is to state that if \(f\in{\mathcal L}_{a,b}'\) and \(\psi\in{\mathcal L}_{c,d}\), with \(a< c,b> d\) then \((f*\psi)(x)= \langle f,\tau_x\check\psi\rangle\) is well defined as a \(C^\infty\) function. Definition 2.6 (without any assumption on \(c\) and \(d\)) contradicts Theorem 2.7 (due to Zemanian) where it is explicitly stated that \(c\leq d\). On page 1357 in the proof of Theorem 2.8, it is claimed that \({\mathcal L}_{c,d}\subset{\mathcal L}_{c,d}'\), which is wrong. For example, \(e^x\in{\mathcal L}_{-1,1}\) but \(e^x\not\in{\mathcal L}_{-1,1}'\), even though the result stated in Theorem 2.8 can be proved by different arguments. The example given in page 1360 for a function in \({\mathcal L}_{-a,-b}\) is not clear. \(f(x)= e^{-|x|}\) is not a \(C^\infty\)-function on \(\mathbb{R}\). If this function is smoothened for \(|x|\leq 1\), then the estimate for \({D^pf()\over k_{a,b}(x)}\) needs a complete revision. The proof of Theorem 3.5 contains several errors. \(T\) is fixed using \(\{a_n\}\) and \(\{b_n\}\). How then can it be assumed that \(b_k- b_n= a_m- a_k= \eta\)? The authors also say that ``the above equalities are assumed whenever \(a_k\) and \(b_k\) are constant sequences'', an absurd statement. Further down the proof, in the same theorem, they choose \(\delta> 0\) depending upon \(q\) and choose \(q\) depending upon this \(\delta> 0\) and both the values of \(q\) are assumed to be the same. For these reasons, the entire proof falls through. In Section 4, it is wrongly stated that the kernel \(G\) is in \({\mathcal L}_{c,d}\) whenever \(c< \alpha_2\) and \(d> \alpha_1\). Indeed, it is \(G(x- t)\) considered as a function of \(t\) which belongs to \({\mathcal L}_{c,d}\) for each fixed \(x\in \mathbb{R}\). In the proof of the inversion theorem on pages 1375--1376, the authors prove that \(g_m=\widehat f_m\) for some \(f_m\in{\mathcal L}_{c,d}'\) and claim that \((f_m* \phi_n)/(\phi_m* \phi_m)\) is a quotient. They should have first proved that the convolution transform is one-to-one (which is true) and should have used the fact that \(X= [{f_n\over\phi_n}]\) is the given Boehmian, so that \([{f_n\over \phi_n}]\) is a quotient. It then follows that \((f_m* \phi_m)/(\phi_m* \phi_m)\) is also a quotient. In Section 6, they give an example of a Boehmian which is convolution transformable but which does not represent an element of \({\mathcal L}_{c,d}'\). Here it is stated that ``\(f_n\) being the uniform limit of a sequence of \(C^\infty\)-functions is itself a \(C^\infty\)-function'', a fact which is not correct. There are plenty of typographical errors (for example, on page 1366 line 14, page 1368 line 1,9, page 1370 line 5 from bottom, page 1371 line 1 from bottom). One should note that this entire theory with better results (and without these errors) is also available in a paper entitled ``Boehmians and their convolution transforms'' by \textit{V. Karunakaran} and \textit{R. Vembu} [Int. J. Math. Game Theory Algebra 13, No. 2, 161--171 (2003; Zbl 1077.46034)]. The authors of the present paper are obviously unaware of this development.