Random probability measures with given mean and variance (Q1866072)

Let \(\mathcal{P}\) denote the set of Borel probability measures on \(\mathbb R,\) let \(\mathcal{P}([0,1])\) be the set of Borel probability measures on \([0,1],\) and let \(\mathcal{P}_{m,V}([0,1])\) be the set of measures in \(\mathcal{P}([0,1])\) with mean \(m\) and variance \(V.\) For \(\mu \in \mathcal{P},\) put \(V(\mu)=\lim\text{Var} \mu .\) A pair \((\mu _{1},\mu _{2})\in \mathcal{P}^{2}\) is said to be a variance split for \(\mu ,\) with splitting probability \(p\in (0,1),\) if \(V(\mu _{1})=V(\mu _{2})<V(\mu)\) and \(\mu =p\mu _{1}+(1-p)\mu _{2}.\) A triangular array \(\{\mu _{n,k},p_{n,k}\}_{n\geq 1,1\leq k\leq 2^{n-1}}\) is said to be a variance split array for \(\mu ,\) if \(\mu =\sum_{k=1}^{2^{n-1}}p_{n,k}\mu _{n,k}, n\geq 1,\) and either \((\mu _{n+1,2k-1},\mu _{n+1,2k})\) is a variance split of \(\mu _{n,k}\) with splitting probability \(p_{n+1,2k-1}/p_{n,k}\) or \(\mu _{n,k}=\mu _{n+1,2k-1}=\mu _{n+1,2k}=\delta _{E\mu _{n,k}}.\) Among the interesting results of this paper, there are the next theorems. Theorem 1. If \(\mu \in \mathcal{P}\) has compact support, then \(\mu \) admits a variance split. Theorem 2. If \(\mu \in \mathcal{P}\) has compact support, then \(\mu \) has a variance split array such that \(\max_{k}V(\mu _{n,k})\rightarrow 0\) as \(n\rightarrow \infty .\) Theorem 3. If \(\{\mu _{n,k},p_{n,k}\}_{n\geq 1,1\leq k\leq 2^{n-1}}\) is a variance split array for \(\mu \) so that \(\max_{k}V(\mu _{n,k})\rightarrow 0\) as \(n\rightarrow \infty ,\) then \(\{E\mu _{n,k}\}\) is dense in the support of \(\mu .\) Now assume \(\mathcal{P}([0,1])\) is endowed with the \(\sigma \)-algebra \(\Sigma\) generated by the weak* topology for \(\mathcal{P}([0,1]).\) The authors use variance split arrays to construct random probabilities \(\mu \in \mathcal{P}_{m,V}([0,1]),\) that is to construct a probability \(P\) on \(\Sigma\) such that \(P(\mathcal{P}_{m,V}([0,1]))=1.\) The paper is very well written, but the numbering (6), (7) and (8) of the corresponding formulae is missing.