What does this mean, you ask? I will give a quote from a fictitious journal article or book. "We now describe our new randomized algorithm for the uniform metrical task system with k states. Recall that every non-negative function from states to the reals is a possible task, but, without loss of generality, we can assume that every task is of the form (0, ... 0,epsilon,0, ... 0), that is, is zero for all but one state, and is arbitrarily small for that one state. ... " Later, in the same fictitious paper, we find: "We now prove that no randomized algorithm for the uniform metrical task system has competitiveness better than the number given in the previous paragraph. Proof: Without loss of generality, we can assume that every task is of the form (0, ... 0,epsilon,0, ... 0), that is, is zero for all but one state, and is arbitrarily small for that one state. ... " Now, think of a class study group reading this article. A member of the group says, "How can they assume that, since it obviously isn't true? They very clearly state that the request can be any vector, not just these special ones! They're contradicting themselves!" Your job is to explain it to that student. One key thing here is that we cannot assume that epsilon is a specific small number. Instead, we can only assume that epsilon is *no larger* than a specific small number, in fact, any small number we name, no matter how small. The way I state it using epsilon and delta is the formally correct way of making that statement, although it never appears in a journal article that way. (It is understood that the reader already knows that it's really epsilon and delta.) Just recall how epsilon and delta are used in a typical proof in elementary calculus.