On transformations of distribution functions on the unit interval -- a generalization of the Gauß-Kuzmin-Lévy theorem (Q687222)

Summary: Let \(X\) be a random variable on [0,1) with the distribution function \(g_ 0\) and \(h\) be a piecewise monotonic transformation on [0,1). Further denote the distribution function of \(h(X)\) by \(g_ 1\), of \(h(h(X))\) by \(g_ 2\), and so on. The distribution functions \(g_ 0,g_ 1,g_ 2,\ldots\) can also be regarded as the asymptotic distribution functions of sequences \((a_ n)\), \((h(a_ n))\), \((h(h(a_ n))),\ldots\). The operator \(P_ h\) defined by \(g_{n+1}'=P_ hg_ n'\) is the so-called Frobenius-Perron operator assigned to the transformation \(h\). We say that \(h\) preserves the distribution function \(k\) if \(g_ 0=k\) implies \(g_ 1=k\) and consequently \(g_ n=k\) for all \(n\), i.e., if \(k'\) is an invariant element of the operator \(P_ h\). Then under weak additional suppositions \(g_ n^{(r)}=k^{(r)}+O(q^ n)\) for \(r=0,1,2\) and for arbitrary initial distribution functions \(g_ 0\), where \(q<1\) depends on \(h\) but not on \(g_ 0\). This result generalizes the Gauss-Kuzmin-Lévy theorem in the metric theory of continued fractions. Particularly, piecewise linear transformations preserve the uniform distribution. In this case more precise estimates are possible under weaker suppositions.