There is irrefutable evidence for some kind of non-locality in quantum physics among entangled particles. See the Bell test page for detailed discussion and examples of the most important paradigm that demonstrates this.
Recent work within the pilot-wave framework has helped to significantly clarify the nature of these non-local interactions, and the broader conflict that they actually pose for all of QM, despite many attempts to downplay these issues from within the standard QM frameworks (Dürr et al., 2014).
At the heart of all of this is a fundamental, pervasive confusion of what this non-locality means, where it comes from within the standard quantum formalisms, and how it might actually work in a more physically plausible, localist manner.
First, there is a widespread confusion that just because you cannot use quantum non-locality to actually communicate arbitrary information at superluminal (faster than light) speeds (Bell, 1975; Ballentine & Jarrett, 1987), that it does not somehow violate the absolute constraints of special relativity. The inability to use this superluminal transmission for communication is due to the fundamental lack of external control we have over these correlations, which must be analyzed “offline” and in aggregate, in comparison with the results from the other observers, and does not contradict the relativistic violation (Norsen, 2011).
However, the tensor-product configuration space representation that is required to compute these non-local measurements is unambiguously incompatible with special relativity (Bell, 1975; Aharonov & Albert, 1981; Peres, 2000; Peres & Terno, 2004; Norsen, 2011). Although this representation can be regarded as a calculational tool and not something that is physically real, we do not have any other such tool or model that can otherwise explain how these results might emerge. Somehow, something seems to be doing some kind of non-local, potentially superluminal signaling (Norsen, 2011; Maudlin, 2011; Shimony, 1993).
A further element of confusion arises because, although the calculational tool is incompatible with special relativity, the results that are computed are not! For example, as shown in Bell, it doesn’t matter what order the putative wave-function collapse occurs within the configuration state space: the same results obtain regardless. Therefore, it appears that the actual physics is still compatible with special relativity, despite the fact that there is no known way to understand how the non-locality arises from an underlying mechanism that does not violate special relativity constraints.
Physical models are local
There is a clear pattern in the examples of calculational tools vs physical models. All of the physical models are based on local propagation of signals according to simple laws, whereas the calculational tools tend to employ non-local equations. This difference is directly tied to the fundamental tradeoffs at work: the calculational tools need non-locality to enable simple single-step calculations, whereas the physical models use local dynamics to enable iterative, autonomous calculations to work in the general case.
For example, the standard QM calculational tools including configuration space, Hilbert space matrix mechanics state vectors, and Fourier space field theory are all fully non-local state representations, and thus cannot help but to produce non-local results. This is analogous to using Newton’s gravitational law or the Coulomb equations for EM: it is baked right into the model.
The fact that physical models depend on local interactions is captured by the notion of a universal basis representation, which identifies the information necessary to compute any possible known physical process. If this information is not present locally, then some kind of non-local communication process would be required, and the logical, information-processing requirements of such a non-local communication process seem to be intractable, requiring some form of routing or lookup tables that would then need to be updated, etc.
Thus, the challenge here is to try to better understand how the underlying physical processes of quantum wave field interactions, unfolding over time through strictly local propagation mechanisms, can end up producing non-local effects consistent with the empirical data. An important available degree of freedom here is that while the speed of light is strictly obeyed by the Maxwell wave equations, it is unclear if such a constraint actually applies to the quantum wave fields.
In the cellular automaton models of quantum wave equations (equivalent to lattice-based wave methods), the speed-of-light factor must be set lower than 1 to achieve numerical stability. Thus, there is no fundamental requirement that everything in such a framework operate at the full speed-of-light update rate of 1 cell per unit time. Therefore, it is entirely possible that the quantum wave function is actually communicated significantly faster.
On balance, the adjustment of an arbitrary update rate seems much less consequential than abandoning the foundational requirement of local mechanisms, which are essential for the overall simplicity of the underlying physical system.
There have been various efforts to explore whether some kind of superluminal process would have other implications that would be in violation of known physics. For example, a recent paper showed that superluminal causal mechanisms should lead to a violation of the no-signalling constraint (Bancal et al., 2012). However, the analysis there did not take into account the role of true randomness (nondeterminism, see stochastic motion), which is essential (Popescu, 2014; Aharonov & Rohrlich, 2008). The non-local correlations in quantum physics are much weaker than theoretically possible, reflecting this randomness. Furthermore, other assumptions may yet produce the observed results in a more naturalistic manner (Reid & Drummond, 2026).
Thus, this whole issue of non-locality remains one of the major open challenges, and it seems premature to draw strong conclusions at this point.