Loop vs. Bernoulli percolation on trees: strict inequality of critical values

We consider loop ensembles on random trees. The loops are induced by a Poisson process of links sampled on the underlying tree interpreted as a metric graph. We allow two types of links, crosses and double bars. The crosses-only case corresponds to the random-interchange process, the inclusion of double bars is motivated by representations of models arising in mathematical physics. For a large class of random trees, including all Galton-Watson trees with mean offspring number in $(1,\infty)$, we show that the threshold Poisson intensity at which an infinite loop arises is strictly larger than the corresponding quantity for percolation of the untyped links. The latter model is equivalent to i.i.d. Bernoulli bond percolation. An important ingredient in our argument is a sensitivity result for the bond percolation threshold under downward perturbation of the underlying tree by a general finite range percolation. We also provide a partial converse to the strict-inequality result in the case of Galton-Watson trees and improve previously established criteria for the existence of an infinite loop in the case were the Galton-Watson tree has Poisson offspring.