A variational formula on the Cramér function of series of independent random variables (Q2358771)

This paper is concerned with the Legendre-Fenchel transform of a cumulant generating function (Cramér function) of a given random variable. The aim is to prove some variational formulas for the Cramér functions of series of independent random variables that depend on the coefficients and Cramér functions of the summands of the series. The cumulant generating function of a random variable \(X\) is defined as \(\psi_X (s)=\ln Ee^{sX}\). It is shown that for each \(t=(t_i)_{i\in I} \in \ell^2\), the cumulant generating function of the random series \[ X_t =\sum_{i\in I} t_i X_i \] is given by \[ \psi_t (s)=\psi_{X_t}(s) =\sum_{i\in I} \psi_i (st_i), \] where \(I\) is a countable set. The main result is the following theorem: Let \((X_i)_{i\in I}\) be a sequence of random variables satisfying certain conditions. Then, for every \(t=(t_i)_{i\in I} \in \ell^2 \), the Cramér function \(\psi^*_{X_t}=\psi^*_t\) of a random series \( X_t =\sum_{i\in I} t_i X_i\) is given by the following variation formula \[ \psi^*_{X_t}(\alpha)=\inf_{b\in \ell^2; \langle t,b\rangle =\alpha} \sum_{i\in I} X^*_i (b_i), \] where \(\psi^*_i\) is the Cramér function of \(X_i\), and \(\alpha \) belongs to an appropriate domain \(D\). The paper is concluded with some examples.