1: \begin{abstract}

2: Given a reversible Markov chain $P_n$ on $n$ states, and another chain $\tilde{P}_n$ obtained by perturbing each row of $P_n$ by at most $\alpha_n$ in total variation, we study the total variation distance between the two stationary distributions, $\| \pi_n - \tilde{\pi}_n \|$. We show that for chains with \textit{cutoff}, $\| \pi_n - \tilde{\pi}_n \|$ converges to $0$, $e^{-c}$, and $1$, respectively, if the product of $\alpha_n$ and the mixing time of $P_n$ converges to $0$, $c$, and $\infty$, respectively. This echoes recent results for specific random walks that exhibit cutoff, suggesting that cutoff is the key property underlying such results. Moreover, we show $\| \pi_n - \tilde{\pi}_n \|$ is maximized by \textit{restart perturbations}, for which $\tilde{P}_n$ ``restarts'' $P_n$ at a random state with probability $\alpha_n$ at each step. Finally, we show that \textit{pre-cutoff} is (almost) equivalent to a notion of ``sensitivity to restart perturbations,'' suggesting that chains with sharper convergence to stationarity are inherently less robust.

3: \end{abstract}

4: