We study the power of measurements implementable with local quantum
operations and classical communication (or LOCC measurements for short) in the
setting of quantum channel discrimination. More precisely, we consider
discrimination procedures that attempt to identify an unknown channel, chosen
uniformly from two known alternatives, that take the following form: (i) the
input to the unknown channel is prepared in a possibly entangled state with an
ancillary system, (ii) the unknown channel is applied to the input system, and
(iii) an LOCC measurement is performed on the output and ancillary systems,
resulting in a guess for which of the two channels was given. The restriction
of the measurement in such a procedure to be an LOCC measurement is of interest
because it isolates the entanglement in the initial input/ancillary systems as
a resource in the setting of channel discrimination. We prove that there exist
channel discrimination problems for which restricted procedures of this sort
can be at either of the two extremes: they may be optimal within the set of all
discrimination procedures (and simultaneously outperform all strategies that
make no use of entanglement), or they may be no better than unentangled
strategies (and simultaneously sub-optimal within the set of all discrimination
procedures).Comment: 17 pages, 4 figure