Maximum efficiency of a linear-optical Bell-state analyzer

In a photonic realization of qubits the implementation of quantum logic is rather difficult due to the extremely weak interaction on the few photon level. On the other hand, in these systems interference is available to process the quantum states. We formalize the use of interference by the definition of a simple class of operations which include linear-optical elements, auxiliary states and conditional operations. We investigate an important subclass of these tools, namely linear-optical elements and auxiliary modes in the vacuum state. For these tools, we are able to extend a previous qualitative result, a no-go theorem for perfect Bell-state analyzer on two qubits in polarization entanglement, by a quantitative statement. We show that within this subclass it is not possible to discriminate unambiguously four equiprobable Bell states with a probability higher than 50%.