A note on a theorem of Karlin (Q5953867)

This mathematical note considers the following problem: Let \(f_0\) and \(f_1\) be two maps from \([0,1]\) into itself, defined by \(f_0(x)= ax\), \(f_1(x)= b+(1- b)x\), where \(0< a,b< 1\), and let \(p\) be a real-valued continuous function on \([0,1]\) with \(0< p(x)< 1\). Suppose a point \(x\in [0,1]\) moves randomly with probability \(p(x)\) to \(f_0(x)\) and with probability \(1-p(x)\) to \(f_1(x)\). This procedure generates a Markov chain on \([0,1]\). The problem is whether the generated Markov chain has necessarily a unique stationary probability measure. This problem is related to a conjecture of S. Karlin (1953). Using a result of \textit{M. Bramson} and \textit{S. Kalikow} [Isr. J. Math. 84, No. 1/2, 153-160 (1993; Zbl 0786.60043)], related also to a paper of A. N. Quas (1996), the author provides an example of place-dependent random iterations with two affine contractions on the unit interval, generating a Markov chain with more than one stationary probability measures. The probability function is continuous and strictly positive. This (counter)example proves that the answer to the above-mentioned problem, as well as to Karlin's conjecture, is negative.