A posteriori error estimates for Markov approximations of Frobenius-Perron operators (Q881609)

Let \((X,\Sigma,\mu)\) be a \(\sigma\)-finite measure space. Let \(L^1\equiv L^1(X,\Sigma,\mu)\) be the Banach space of all \(\mu\)-integrable real functions defined on \(X\) with the \(L^1\)-norm \(\| f\|= \int|f|\,d\mu\). Let \(S: X\to x\) be a nonsingular transformation, i.e. \(\mu(S^{-1}(A))= \mu(A)\) for all \(A\in X\) such that \(\mu(A)= 0\). The unique operator \(P\equiv P_S: L^1\to L^1\) defined by \(\int_A Pf\,d\mu= \int_{S^{-1}(A)} f\,d\mu\) for all \(A\in\Sigma\), is called the Frobenius-Perron operator corresponding to \(S\). Since \(P\) is a positive linear operator and preserves the integral, a Frobenius-Perron operator is a Markov operator. Given a density function \(f\), the probability measure given by \(\mu_f(A)= \int_A f\,d\mu\) for all \(A\in\Sigma\) is invariant under \(S\) if and only if \(f\) is a fixed point of \(P\) (the invariance of \(\mu_f\) means that \(\mu_f(S^{-1}(A))= \mu_f(A)\) for all \(A\in\Sigma\)). A fixed density \(f^*\) of a Frobenius-Perrron operator \(P\) is a fixed point of \(P\) which is a density function (i.e. \(f^*\) is nonnegative with \(L^1\)-norm 1). The authors consider a class of quasi-compact Frobenius-Perron operators. Let \(X= \Omega\) with the usual Lebesgue measure, where \(\Omega\subset\mathbb{R}^N\) is a bounded region, and let \(BV\equiv BV(\Omega)\) be the Banach space of functions of bounded variation in \(L^1\) with norm \(\| f\|_{BV}= V(f)+\| f\|\), where \(V(f)\) is the variation of \(f\) over the domain. Then \(P: (BV,\|\;\|_{BV})\to (BV,\|\;\|_{BV})\) is quasi-compact if there is a positive integer \(r\) and a compact operator \(K:(BV,\|\;\|_{BV})\to (BV,\|\;\|_{BV})\) such that \(\| P^r- K\|_{BV}< 1\). For the introduced quasi-compact Frobenius-Perron operators it is shown, if \(\{f_n\}\subset BV\) is a sequence of approximation to a fixed density \(f^*\in BV\), then the residual \(\| f_n- Pf_n\|_{BV}\) is equivalent to the exact error \(\| f^*- f_n\|_{BV}\) even though we do not know any information about the exact fixed density \(f^*\). The given numerical results also indicate that the quantity \(\| f_n- Pf_n\|_{BV}\) is indeed a very good indicator of the exact error \(\| f^*- f_n\|_{BV}\).