Biorthogonal analogue of the Hermite polynomials and the inversion of the Fourier transform with respect to a non-Gaussian measure (Q1814369)

Consider the Hilbert space \(H\) and the separable Banach space \(B\) such that the dense Hilbert-Schmidt embeddings \(B^*\subset H \subset B\) take place. Let \(\mu\) be a \(\sigma\)-additive measure on \(B\) with a smooth characteristic function \(\chi(\theta)= \int_ B e^{i(\theta,x)}\mu(dx)\) and logarithmic gradient \(\lambda(x)\) \((x \in B)\) defined by a formula of integration by parts of \[ \int_ B\langle h,f'(x)\rangle\mu(dx)=-\int_ Bf(x)\langle h,\;\lambda(x)\rangle\mu(dx) \] \((h \in B^*\), \(f\) a smooth function on \(B)\). A set of polynomials \[ P(\varphi_ 1,\dots,\varphi_ n,x)=(-i)^ n \left\langle \varphi_ 1, {\partial \over \partial \theta} \right\rangle \cdots \left\langle \varphi_ n,{\partial \over \partial \theta} \right\rangle |_{\theta=0}{e^{i \langle \theta,x \rangle} \over \chi(\theta)} \] and a set of functions \(Q(\varphi_ 1,\dots,\varphi_ n;x)=\left\langle \varphi_ 1,\lambda(x)+{\partial \over \partial x} \right\rangle \cdots \left\langle \varphi_ n,\lambda(x)+{\partial \over \partial x}\right\rangle\cdot 1\) are defined. These systems are biorthogonal \[ \int_ BP(\varphi_ 1,\dots,\varphi_ n;x)Q_ m(\psi_ 1,\dots,\psi_ m;x)\mu(dx)=\delta_{n_ m} \sum_{(k_ 1,\dots,k_ n)} \prod^ n_{s=1}(\varphi_ j,\psi_{k_ j}). \] In the case of Gaussian measure \(\mu\) both systems coincide with Hermite polynomials (within normalisation). The constructed system of polynomials has some properties similar to the ones of Hermite polynomials. The transformation \(f\mapsto \hat f(\theta)=\int_ B{e^{i \langle\theta,x \rangle}\over \chi(\theta)} f(x) \mu(dx)\) is considered. In the Gaussian case it coincides with the Fourier-Wiener transform. The inverse transformation is constructed by the decomposition with respect to the system \(\{Q\}\). Notice that the questions considered there are connected with the notion of the extended stochastic integrals.