Generalized integrated Cauchy functional equation with applications to probability models (Q2241714)

Let $H(x):=\int H(x+y)\mu(dy)$, $x\in S$, where $S$ be an abelian topological semigroup with zero element, $\mu$ be a nonnegative measure defined on $S$ with $\mu(\{0\})<1$, and $H$ be a nonnegative real-valued Borel measurable function on $S$, with $H(0)>0$. The author studies the problem of solving a generalized integrated Cauchy functional equation (ICFE referred to (*)). Special cases of this equation have applications to various stochastic or statistical models, referred to in this article as probability models. Specialized versions of ICFE or their variations have been studied by \textit{G. Choquet} and \textit{J. Deny} [C. R. Acad. Sci., Paris 250, 799--801 (1960; Zbl 0093.12802)], \textit{P. L. Davies} and \textit{D. N. Shanbhag} [Q. J. Math., Oxf. II. Ser. 38, 13--34 (1987; Zbl 0617.60016)], \textit{D. N. Shanbhag} [J. Appl. Probab. 14, 640--646 (1977; Zbl 0375.60021); Comput. Stat. Data Anal. 12, No. 1, 115--126 (1991; Zbl 0825.62360)], \textit{K.-S. Lau} and \textit{C. R. Rao} [Sankhyā, Ser. A 44, 72--90 (1982; Zbl 0584.62019)], \textit{P. Ressel} [Ann. Probab. 13, 898--922 (1985; Zbl 0579.60012)], \textit{C. R. Rao} and \textit{D. N. Shanbhag} [Adv. Appl. Probab. 18, 660--678 (1986; Zbl 0607.62005); Q. J. Math., Oxf. II. Ser. 40, No. 159, 333--350 (1989; Zbl 0695.60021); J. Multivariate Anal. 38, No. 1, 141--148 (1991; Zbl 0732.60009); Handb. Stat. 19, 733--763 (2001; Zbl 0983.60024)], \textit{K.-S. Lau} and \textit{W.-B. Zeng} [Stud. Math. 97, No. 2, 115--135 (1990; Zbl 0719.43002)] and \textit{G. J. Székely} and \textit{W.-B. Zeng} [J. Theor. Probab. 3, No. 2, 361--365 (1990; Zbl 0706.60006)]. Using technical lemmas (Lemma 2.1; Lemma 2.2; Lemma 2.3) the author shows the following results the underlying assumption in these is that (*) is met, additionally, with $H$ continuous and satisfying $H(0)=1$, and $\mu$ so that $\mu((\mathrm{supp}[\mu])^{c})=0$. More precisely, using auxiliary results proven in several lemmas, the author shows the following main results: Theorem 1: Let $S^*$ and B be as in Lemma 2.1 (see the paper) and, for, each $x\in S^*$, let $\xi(x,.)$ be defined as follows: $\xi(x,\omega)=\{\lim\xi_{n}(x,\omega)$ if the limit exists and is finite 0 otherwise, where $\xi_{n}(x,\omega)=H(x+Y_{n}(\omega))/H(Y_{n}(\omega))$ if $H(Y_{n}(\omega))>0$ and $H(x)>0$, 0 otherwise, on the probability space, \[ P_{H}(B_1\times B_2\times\dots B_{j}\times S_2^{N_{j}})=(H(0))^{-1}\int\dots\int H\left(\sum x_{i}\right)\left(\prod\mu_2(dx_{i}\right), \eqno{(\ominus)} \] where $N_{j}=\{j+1,j+2,\dots\}$. Then, in familiar notation, with $B=\{x\in S: H(x)>0\}>\eta(x,.)=0$, $x\in B^{c}$; $E_{H}(\eta(x,.))=H(x)$, $x\in S^*\cup B^{c}$; $\int\eta(x,.)\mu(dx)=1$ a.s. $[P_{H}]$, and $\eta(x_1,.)\eta(x_2,.)=\eta(x_1+x_2,.)$ a.s. $[P_{H}]$, $(x_1,x_2)\in (S^{*}\cup B^{c})\times S^{*}$. Theorem 2: Suppose, that the assumptions of the previous theorem are met, and, unless $S^{*}=S$, for each $x\in(S^{*})^{c}$, one has a $\sigma$-finite measure $\nu_{x}$ such that $\nu_{x}(\{z\in S: x+z\in(S^{*})^{c}\})=0$, and $H(x+y)=\int H(x+y+z)ν_{x}(dz)$, $y\in S$. Then, there exists a nonnegative Borel measurable function $e$ on $S\times\Omega$ (in standard notation), such that, for each $x\in S^{*}$, $e(x,.)$ equals $\eta(x,.)$ with notation $e$ in place of $\eta$ and, respectively, $S$ in place of $S^*\cup B^{c}$ and $S^2$ in place of $(S^*\cup B^{c})\times S^*$, are valid; with notation $\eta$ changed to $e$, are valid, in this case, is obvious. Theorem 3: Suppose the assumptions of Theorem 2 are met, with $S$ of the form $\prod S_{i}$ having, for each $i$, $S_{i}=Z$ or $N_0$ or $-N_0$ or $R$ or $R_+$ or $-R_+$, $H$ continuous (w.r.t. the concerned topology), and $\lambda$ be the restriction to $S$ of a Haar measure defined on the smallest subgroup of $R^n$ generated by $S$. Let $h:S\to R_+$ be a Borel measurable function that is locally integrable with respect to $\lambda$, and $\mu$ be a $\sigma$-finite measure on $S$ such that $\mu(\{0\})<1$ and \[ H(.)=\int H(.+y)\mu(dy),\text{ a.e. }[\lambda]. \] Also, unless $S^*=S$, let, for each $x\in (S^*)^{c}$, there exists a $\sigma$-finite measure $\nu_{x}$ on $S$, such that $\nu_{x}(\{z\in S: x+z\in(S^*)^{c}\})=0$ and \[ H(x+.)=\int H(x+z+.)ν_{x}(dz),\text{ a.e. }[\lambda]. \] Then \[ H(.)=\int\exp\{<x,y>\}\alpha(dx),\text{ a.e. }[\lambda], \] where $\alpha$ is a measure on $([-\infty,\infty])^n$, such that, for a.a. $[\alpha]x$, $\int\exp\{<x,y>\}\mu(dy)=1$ and $\exp\{<.,x>\}$ is well defined a.e. $[\lambda]$. Theorem 4: Let the assumptions in Theorem 1 be met, in conjunction with that in this case $S$ is a Polish space. Then, de Finetti's theorem applies to the sequence, $\{X_{n}\}$, of exchangeable random elements defined on the probability space satisfying $(\ominus)$, and, for each $x\in\mathrm{supp}[\mu_2]$, the sequence $\{\eta_{n}(x,.):n=1,2,\dots\}$,with, for each $n$, $\eta_{n}(x,.)=P_{H}\{X_1\in O_{n}(x)|\mathfrak{I}\}/\mu_2(O_{n}(x))$, converges in $L_2$ (or, indeed, in $L_\alpha$ for any $\alpha\geq 1$) to $\xi(x,.)$, where $\mathfrak{I}$ is the invariant $\sigma$-field relative to the sequence $\{X_{n}\}$, and $\{O_{n}(x)\}$ is the sequence of open spheres, with centre $x$ and radii $(1/n)$. (That some subsequence of $\{\eta_{n}(x,.)\}$ converges to $\xi(x,.)$ a.s. $[P_{H}]$ is, hence, obvious). The popularity of the subject is due to its high degree of importance not only within mathematics but also in a wide range of applications especially in damage models, extreme value distributions and some miscellaneous problems.