Uniqueness of embedding of Gaussian probability measures into a continuous convolution semigroup on simply connected nilpotent Lie groups (Q943647): Difference between revisions

Let \(G\) be a simply connected nilpotent Lie group, let \(\left(\mu_t\right)\) denote a continuous convolution semigroup of probabilities on \(G\) with generating functional \(A:= \frac{d^+ }{dt }|_{t=0}\mu_t\) (defined on \(C^\infty_b(G)\)). Via the Lévy Khinchin representation \(A\) resp. \((\mu_t)\) is characterized by a triple consisting of a left invariant differential operator of order 1, the drift, a non-negative definite matrix (representing the Gaussian part) and a non-negative measure \(\eta\), the Lévy measure. \((\mu_t)\) resp. \(A\) is Gaussian, iff \(\eta =0\). A challenging problem still not completely solved in probabilities on groups is the problem of uniqueness of embedding: \(\mu\) is called embeddable, if there exists a continuous convolution semigroup \(\left(\mu_t\right)\) with \(\mu = \mu_1\); \(\mu\) is called uniquely embeddable, if for continuous convolution semigroups \(\left(\mu_t^{1}\right)\) and \(\left(\mu_t^{2}\right)\) with \(\mu =\mu_1^{(1)} = \mu_1^{(2)}\) follows \(\mu_t^{(1)} = \mu_t^{(2)}\) for all \(t\geq 0\). Up to now only partial results are known, e.g. uniqueness of semistable embedding [cf., e.g., \textit{W. Hazod, E. Siebert}, Stable Probability Measures on Euclidean Spaces and on Locally Compact Groups. Structural Properties and Limit Theorems. Mathematics and its Applications (Dordrecht). 531. Dordrecht: Kluwer Academic Publishers. (2001; Zbl 1002.60002)], uniqueness of embedding into Gaussian semigroups on step 2 nilpotent groups [cf. e.g., the author, Publ. Math. 53, No. 3--4, 415--422 (1998; Zbl 0916.60010); \textit{G. Pap}, Arch. Math. 62, No. 3, 282--288 (1994; Zbl 0804.60007)]. In this note the author proves uniqueness of embedding for Gaussian laws on general simply connected nilpotent Lie groups. Precisely, in Theorem 1 it is shown, that if \(\mu =\mu_1^{(1)} = \mu_1^{(2)}\) and \(\mu_t^{(1)}\) is Gaussian, then \(\mu_t^{(1)} = \mu_t^{(2)}\) for all \(t\geq 0\).