Rigorous computation of linear response for intermittent maps (Q6140926)

The authors study a rigorous computation of linear response for intermittent maps. They consider the maps \(T_{\alpha}(x)= x (1+2^{\alpha} x^{\alpha})\) if \(x\in [0, 1/2]\), \(T_{\alpha}(x)= 2x-1\) if \(x\in (1/2, 1]\). Let \(\alpha_{0}=1/8\) and denote by \(h_{\epsilon}\) the \textit{a.c.i.p.} of \(T_{\alpha_{0}+\epsilon}\). First they provide assumptions on the dynamical system and state some consequences. Overall, the obtained results extend those from previous studies on intermittent maps away from the boundary (possibly really far from it) with remarkable computing power, thus allowing to compute the linear response for LSV maps as the exponent at the indifferent fixed point changes. Let us recall that a map \(S : [0, 1] \rightarrow [0, 1]\) is said to be LSV if: (i) \(S\) is nonsingular with respect to Lebesgue measure \(m\); (ii) \(S\) has two onto branches, \(S_{0} : [0, 0.5) \rightarrow [0, 1)\), and \(S_{1} : [0.5, 1] \rightarrow [0, 1]\); (iii) \(S_{0}\) and \(S_{1}\) have inverses \(g_{0}\) and \(g_{1}\) respectively, and \(g_{i}\in C^{3}\) for \(i = 0, 1\). Denote by \(L_{\epsilon}\) the Perron-Frobenius operator associated with \(T_{\epsilon}\), i.e., for \(\varphi \in L^{\infty}\) and \(\psi\in L^{1}\) one has \[ \int \varphi \circ T_{\epsilon}\cdot\psi\, dm = \int \varphi \cdot L_{\epsilon}\psi \, dm. \] The authors provide answers to the following three questions for general intermittent maps: (1) How to approximate efficiently the involved discretized operators? (2) How to bound the approximation errors involved in the discretization? (3) How to bound explicitly and efficiently the constants used in the proofs? The authors study properties of the induced map, and its relationship with the original map, in particular discussing some theoretical aspects behind the approximation of the density for the induced map. They also mention Chebyshev projection and convergent rates. Further, they describe the approximation of the linear response for the induced map, providing the error in approximating the linear response. Furthermore, they pull back the measure to the original map and normalize the density. Moreover, the authors compute an approximation with an explicit error of the linear response for an LSV map. They provide numerical remarks based on the ``shooting method'' and consider automated Lasota-Yorke inequalities. Effective bounds for the all constants involved in the proofs are given. The authors mention that their algorithm allows them a rigorous control of all the numerical and discretization errors. They also discuss the technique to compute some of the functions involved in their approximationa in Appendix A.