6
$\begingroup$
  • Start with a bond-percolation process just above criticality, say $p=1/2+\varepsilon$ on the graph $\mathbb Z^2$ with $\varepsilon>0$.
  • Sample $D\in\{0,1\}^E$ from an independent product measure with $\mathbb P(D_e=1)=\delta>0 \;\forall e\in E$.
  • Remove all edges $e$ with $D_e=1$ from the percolation process.

Clearly one ends up with another percolation process with parameter $p'=(1/2+\varepsilon)(1-\delta)<1/2$, i.e. below criticality (for $\varepsilon$ sufficiently small). In particular the resulting process has no infinite connected components almost surely.

Now what happens if the process $D$ of edges to be removed is not a product measure (but still essentially translation invariant and such that a positive fraction of edges are deleted)? Can we still choose $\varepsilon$ sufficiently small so that there won't be any infinite connected components after deleting these edges from our original percolation process?

(Partial answer) In case the law of $D$ stochastically dominates some (non-degenerate) Bernoulli product measure, the same reasoning as before applies.

(The problematic case) What if the law of $D$ is strongly "repellant"? An extreme example would be a law of $D$ where deleting a pair of adjacent edges has probability $0$, i.e. $\mathbb P(D_e=D_{e'}=1)=0$, whenever $e,e'$ share a vertex.

Edit: $D$ should still be independent of the original percolation process.

$\endgroup$
0

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.