Reading some papers on security amplification of weak-PRP written by Stefano Tessaro, I noticed the use of notion "epsilon-PRP" and the notion of "security amplification". These notions, in conjunction with the statement
We consequently call a primitive satisfying this assumption an epsilon-PRP: Clearly, AES is much more likely to be a 0.99-PRP, rather than a fully-secure PRP.
Can anybody explain when security of PRP is higher, for epsilon very close to 1? Security amplification of cascade of PRP's occur when resulting epsilon is greater than every epsilon of basic PRP?