Bell violation with path-entangled number states under realistic detection

Abstract

A single photon delocalised over two modes is entangled, and whether that entanglement can violate a Bell inequality has been disputed for three decades. In principle it is settled: the surviving objection was the absence of a shared phase reference, and a local oscillator supplies one, acting as a shared frame rather than as a measurement setting. In practice it is open, and what remains is a question about the apparatus. Loss does not spoil a parity measurement so much as rescale it, carrying it into the same one-parameter family of detector operators that already contains on--off detection. Within that family we give closed correlators for path-entangled number states, or N00N states, at arbitrary efficiency, with dark counts, mode mismatch, phase noise and upstream loss. Because loss and displacement commute up to a rescaling of the displacement amplitude, loss before and after the displacement is interchangeable, and a symmetric experiment is governed by one overall efficiency: the probability that a heralded photon is detected. For a single photon the Clauser--Horne--Shimony--Holt threshold on that efficiency is 0.83 with on--off detection and 0.95 with parity, so the scheme that needs no photon-number resolution is the more robust, by twelve percentage points of loss and a factor of five in tolerable dark counts. Efficiency is no longer the obstacle; the mode overlap is. The best demonstrated telecom coupling, a buildable interferometer and today's detectors give an overall efficiency of 0.86, at which the required overlap is 0.92. We show that this overlap carries the heralded photon's single-mode weight as well as the local oscillator's mode match, so buying more of it costs heralding efficiency --- and no source has reported the two together.