Remarks on the \(r\) and \(\Delta\) convolutions (Q2491119): Difference between revisions

Let \(A_{i}\) be C\(^{\ast}\)-algebras with \(1\), \(i=1,\dots ,N\), \(N\) possibly \(\infty\), \(\varphi_{i}\) and \(\psi_{i}\) be states on \(A_{i}\), \(A\) the algebraic free product of \(A_{i}\). The conditional free product state \(\Phi\) on \(A\) is that satisfying \(\Phi (a_{i_{1}}\cdots a_{i_{m}})=\varphi (a_{i_{1}})\cdots\varphi (a_{i_{m}})\) for \(i_{k}\neq i_{k+1}\), \(a_{i}\in A_{i}\) and \(\psi_{i_{k}}(a_{i_{k}})=0\). It is considered together with the free product of the \(\psi_{i}\). If \(X_{i}\) has distributions \(\mu_{i}\), \(\nu_{i}\) with respect to \(\varphi_{i}\), \(\psi_{i}\), respectively, then the distribution of \(X_{1}+\dots +X_{i}+\dots \) with respect to \(\Phi\), together with the additive free convolution of the \(\nu_{i}\), is the conditionally free convolution of \((\mu_{i},\nu_{i})\). The \(r\)-free convolution of \(\mu_{i}\) is the above for \(\nu_{i}=r\mu_{i}+(1-r)\delta_{0}\). It is associative if and only if \(r=0\) or \(1\) (Theorem 4). Theorems 1, 2. If \(\mu\) has compact support, mean \(0\) and variance \(1\), then, for \(N\rightarrow\infty\), \(r<1\), the \(r\)-free convolution of \(N\) measures all equal to \(D_{_{N}-1/2}\mu\) converges \(\ast \) -weakly to the ``Kesten measure \(k_{r}\)'', explicitly defined. With \((1-\lambda N^{-1})\delta_{0}+\lambda N^{-1}\delta_{1}\) instead of \(D_{_{N}- 1/2}\mu\), the limit is a pair involving conditional free Poisson measures, also explicitly defined in the paper. Theorem 3. Let \({\mathcal K}\) be a real \(N\)-dimensional Hilbert space, \(N<\infty\), \({\mathcal K}_\mathbb{C}\) its complexification, \({\mathcal F}({\mathcal K})=\mathbb{C}\Omega\oplus\oplus_{n\geq 1}\otimes^{n}{\mathcal K}_\mathbb{C}\) with scalar product multiplied by \(r^{n-1}\) on the \(n\)-th component, \(e_{1},\dots ,e_{N}\) an orthonormal basis in \({\mathcal K}\), \(a_{i}\in M_{n}(\mathbb{C})\), \(c (e_{i}) (x_{1}\otimes\cdots\otimes x_{n})=r \langle e_{i},x_{1}\rangle x_{2}\otimes\cdots\otimes x_{n}\), \(c (e_{i})\Omega = 0\), \(\omega_{r}(e_{i})=c(e_{i})^{\ast}+c(e_{i})\). Then, for \(r>0\), \[ \left\|\sum a_{i}\otimes c (e_{i})\right\| =\max(1,r^{1/2})\left\|\sum a_{i}^{\ast}a_{i}\right\|^{1/2}, \] \[ \left\|\sum a_{i}\otimes\omega_{r}(e_{i})\right\|\in [b,2\max(1,r^{1/2})b], \] \(b=\max(\|\sum a_{i}^{\ast}a_{i}\|^{1/2},\|\sum a_{i}a_{i}^{\ast}\|^{1/2})\), while the last expression, for \(r=0\), equals \(b\). Let \(\varphi_{i}\) be states on the \(\ast \)-algebras \(A_{i}\) without \(1\), \(A_{i}^{\sim}=A_{i}\oplus\mathbb{C}\cdot 1\), \(\varphi_{i}^{\sim}(x+\alpha\cdot 1)=\varphi_{i}(x)+\alpha\), \(\delta_{i}(x+\alpha\cdot 1)=\alpha\), \(\psi_{i}=r\varphi_{i}^{\sim}+(1- r)\delta_{i}\). The \(r\)-free product of \(\varphi_{i}\) is defined as the restriction of the conditionally free product of \((\varphi_{i}^{\sim},\psi_{i})\) to the free product of \(A_{i}\). It is associative if and only if \(r=0\) or \(1\) (Theorem 5). The \(\Delta\)-convolution is defined by considering a \(\nu_{i}\) having as moments those of \(\mu_{i}\) (\(k\geq 1\)) multiplied by the moments of a fixed probability measure \(\Delta\). It is associative (Theorem 7) if and only if the moments of \(\Delta\) are \(a^{k}\) with \(a=0\) or \(1\). The \(r^{\ast}\)- convolution of measures is defined by addition of certain cumulants. It is positive for all pairs of probability measures if and only if \(r=0\) or \(1\) (Theorem 6). The paper contains three open problems: defining the convolution in the central limit theorem for \(r>1\), best estimation in the proved inequalities, and defining a \(\Delta\ast\) convolution by adding the cumulants, which leads to probability measures when applied to such.