WorksPhysics and Astronomy
Randomized routing: ruling out classical networks without trusting source independence
Abstract
Network nonlocality without inputs assumes that the sources are independent, but no data can test this, since one classical source feeding all parties reproduces any correlation. Here we propose a randomized routing strategy: a source is duplicated, and the two parties it connects choose at random which copy to read. Whether any classical network, even with correlated sources, explains the routed data is then a linear program, and its infeasibility certificates are linear inequalities on the data. Source independence is thus traded for locality and free choice of the routing, which experiments can enforce. In the triangle scenario, the RGB4 distribution fails the test over a wide range of parameters, including a point where its triangle data alone have a numerically found model with independent classical sources, and it still fails with 3% white noise at each source.