In the setting of concurrent self composition, a single protocol is executed many times concurrently by a single set of parties. In this paper, we prove that there exist many functionalities that cannot be securely computed in this setting. We also prove a communication complexity lower bound on protocols that securely compute a large class of functionalities in this setting. Specifically, we show that any protocol that computes a functionality from this class and remains secure for m concurrent executions, must have bandwidth of at least m bits. Our results hold for the plain model (where no trusted setup phase is assumed), and for the case that the parties may choose their inputs adaptively, based on previously obtained outputs. While proving our impossibility result, we also show that for many functionalities, security under concurrent self composition (where a single secure protocol is run many times) is actually equivalent to the seemingly more stringent requirement of security under concurrent general composition (where a secure protocol is run concurrently with other arbitrary protocols). This observation has significance beyond the impossibility results that are derived by it for concurrent self composition.
Postscript, gzipped Postscript.