0% found this document useful (0 votes)
3 views7 pages

Test of The Physical Signi Cance of Bell Non-Locality: Article

The article discusses the implications of loophole-free violations of Bell inequalities, demonstrating that if any assumption of local hidden-variable theories fails, it must fail completely. It presents two key results: quantum correlations cannot be simulated with hidden-variable theories that assume outcome independence but allow for partial measurement dependence, and vice versa. The findings challenge the prevalent view that resolving which assumptions fail requires metaphysical judgment, suggesting instead that it can be approached through experimental means.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
3 views7 pages

Test of The Physical Signi Cance of Bell Non-Locality: Article

The article discusses the implications of loophole-free violations of Bell inequalities, demonstrating that if any assumption of local hidden-variable theories fails, it must fail completely. It presents two key results: quantum correlations cannot be simulated with hidden-variable theories that assume outcome independence but allow for partial measurement dependence, and vice versa. The findings challenge the prevalent view that resolving which assumptions fail requires metaphysical judgment, suggesting instead that it can be approached through experimental means.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 7

Article https://doi.org/10.

1038/s41467-025-59247-7

Test of the physical significance of


Bell non-locality
1,2 1 3,4
Received: 15 February 2025 Carlos Vieira , Ravishankar Ramanathan & Adán Cabello

Accepted: 16 April 2025

Loophole-free violations of Bell inequalities imply that at least one of the


assumptions behind local hidden-variable theories must fail. Here, we show
Check for updates
that, if only one fails, then it has to fail completely, therefore excluding
models that partially constrain freedom of choice or allow for partial retro-
1234567890():,;
1234567890():,;

causal influences, or allow partial instantaneous actions at a distance. Spe-


cifically, we show that (i) any hidden-variable theory with outcome
independence (OI) and arbitrary joint relaxation of measurement indepen-
dence (MI) and parameter independence (PI) can be experimentally exclu-
ded in a Bell-like experiment with many settings on high-dimensional
entangled states, and (ii) any hidden-variable theory with MI, PI and arbitrary
relaxation of OI can be excluded in a Bell-like experiment with many settings
on qubit-qubit entangled states.

In the early days of quantum theory, the question of whether there is p(a, b∣λ, x, y) and p(λ∣x, y) so
deeper theory underlying quantum theory was considered “a philo- Z
sophical question for which physical arguments alone are not pða, bjx, yÞ = dλ pðλjx, yÞpða, bjλ, x, yÞ: ð1Þ
decisive”1. Bell’s theorem2,3 made it possible to exclude experimentally
some of these deeper theories, called hidden-variable (HV) theories4. (1) Measurement independence (MI): For every pair of particles, the
Today, Bell tests5–9 have convinced us that some HV theories cannot measurements (x, y) are not correlated with λ. That is, p(x, y∣λ) = p(x, y),
explain what we see. which, through Bayes’s theorem, is equivalent to
In a Bell test, a source of pairs of particles sends each particle to
pðλjx, yÞ = pðλÞ: ð2Þ
a different laboratory. In the first laboratory, an observer (Alice)
chooses to measure x ∈ X and obtains a ∈ A. In the second labora- Therefore, the knowledge of λ gives no information about (x, y), and
tory, a different observer (Bob) chooses to measure y ∈ Y and vice versa.
obtains b ∈ B. After many repetitions, Alice and Bob compute the (2) Outcome independence (OI)11 also referred to as completeness10,12:
joint probability of (a, b) given (x, y), denoted p(a, b∣x, y). The set p(a∣λ, x, y, b) is independent of b, and hence may be written
{p(a, b∣x, y)}x∈X,y∈Y,a∈A,b∈B is called a correlation for the Bell scenario
pðajλ, x, y, bÞ = pðajλ, x, yÞ: ð3Þ
(∣X∣, ∣A∣; ∣Y∣, ∣B∣), in which Alice can choose between ∣X∣ measurement
settings with ∣A∣ possible outcomes and Bob between ∣Y∣ settings
with ∣B∣ outcomes. Similarly, p(b∣λ, x, y, a) = p(b∣λ, x, y).
Bell’s theorem asserts that no HV model satisfying some (3) Parameter independence (PI)11, initially called locality10:
assumptions can reproduce certain quantum correlations. These p(a∣λ, x, y) is independent of y, and hence may be written as
models are collectively called “local” HV models and are defined as
pðajλ, x, yÞ = pðajλ, xÞ: ð4Þ
those satisfying the following assumptions10,11:
(0) Hidden variables. There are hidden variables that associate to
each pair of particles a state λ ∈ Λ and underlying probability densities Similarly, p(b∣λ, x, y) = p(b∣λ, y).

1
Department of Computer Science, The University of Hong Kong, Pokfulam Road, Hong Kong, Hong Kong. 2Instituto de Matemática, Estatística e Computação
Científica, Universidade Estadual de Campinas, CEP 13083-859 Campinas, Brazil. 3Departamento de Física Aplicada II, Universidad de Sevilla, E-41012
Sevilla, Spain. 4Instituto Carlos I de Física Teórica y Computacional, Universidad de Sevilla, E-41012 Sevilla, Spain.
e-mail: carlos.humberto.vieira@outlook.com; ravi@cs.hku.hk

Nature Communications | (2025)16:4390 1


Article https://doi.org/10.1038/s41467-025-59247-7

Assumptions (2) and (3) are independent10 and, together, imply signaling correlation can be reproduced with an HV model with
that complete MD. The complete relaxation of MI using alternative ways of
quantifying MD14,18 matches the above definition of complete MD (see
pða, bjλ, x, yÞ = pðajλ, xÞpðbjλ, yÞ, ð5Þ also Supplementary Note1).

which is Bell’s original assumption2 which is now called local factoriz- Quantifying parameter dependence. A correlation p(a, b∣x, y) can be
ability or local causality. reproduced with an (εA, εB)-parameter dependent [(εA, εB)-PD] HV
Assumption (0) is the expression of the belief in a deeper theory model14 if it can be reproduced with an HV model such that, for all x,
underlying quantum theory. MI is motivated by the assumption that y, y′ (y, x, x′), and for all λ,
each of the observers has freedom of choice13 or, more generally, by
the assumption that which specific measurements are actually per- 1X
jpðajλ, x, yÞ  pðajλ, x, y0 Þj ≤ εA , ð7aÞ
formed is not governed by the HVs that govern the particles. OI is 2 a
based on the assumption that, as it happens in deterministic models
[i.e., when p(a, b∣λ, x, y) ∈ {0, 1}], if we would know λ, we would
1X
observe that p(a, b∣λ, x, y) = p(a∣λ, x, y)p(b∣λ, x, y)14. PI is grounded on jpðbjλ, x, yÞ  pðbjλ, x 0 , yÞj ≤ εB : ð7bÞ
2 b
the assumption that superluminal signalling between one party’s
choice and the other party’s spacelike separated outcome is
impossible10. Therefore, p(a, b∣x, y) can be reproduced with an HV model with PI if,
Existing experiments are inconclusive about which assumptions and only if, it can be reproduced with an (εA, εB)-PD HV model with
fail. As a consequence, possible explanations include HV theories with εA = εB = 0. If not, we say that p(a, b∣x, y) can be reproduced with an HV
different degrees of “measurement dependence”12,14–21 (that may occur model with partial PD if it can be reproduced with an (εA, εB)-PD HV
due to limitations to freedom of choice22,23 or to retrocausal model for some 0 < εA, εB < 1. Finally, p(a, b∣x, y) can only be reproduced
influences24,25), different amounts of instantaneous “actions at a with HV models with complete PD if p(a, b∣x, y) cannot be reproduced
distance”26–32, and combinations thereof33. At least one of the four with any (εA, εB)-PD HV model for any εA, εB < 1. Any non-signaling
assumptions is false. But which one or which ones?11[pp. 124, 149, correlation can be reproduced with HV models with complete PD.
96],34[p. 66]. The prevalent view is that advancing in the resolution of
this problem is not possible “on purely physical grounds but it requires Quantifying outcome dependence. A correlation p(a, b∣x, y) can be
an act of metaphysical judgement”35. Here, we challenge this view and reproduced with a δ-outcome dependent (δ-OD) HV model14 if it can be
present two results. Result 1 shows that there are quantum correlations reproduced with an HV model such that, for all x, y, a, a′, and for any λ,
that cannot be simulated with any HV theory assuming OI but partial
(as opposed to complete) measurement dependence (MD) or partial 1X
∣pðbjλ, x, y, aÞ  pðbjλ, x, y, a0 Þ∣ ≤ δ: ð8Þ
parameter dependence (PD). Result 2 shows that there are quantum 2 b
correlations that cannot be simulated with any HV theory assuming MI,
PI but partial outcome dependence (OD). Therefore, p(a, b∣x, y) can be reproduced with an HV model with OI if,
In Sec. II A, we introduce the standard ways to quantify MI, PI and and only if, it can be reproduced with a δ-OD HV model with δ = 0. We
OI. Result 1 is presented in Sec.II B, where we also describe an say that p(a, b∣x, y) can be reproduced with an HV model with partial
experiment to exclude HV theories with partial MD and PD. Result 2 is OD if it can be reproduced with a δ-OD HV model with 0 < δ < 1. We say
presented in Sec. II C, which includes the description of an experiment that p(a, b∣x, y) can only be reproduced with an HV model with com-
to exclude HV theories with partial OD. The consequences and appli- plete OD if it cannot be reproduced with any δ-OD HV model for any
cations of the results are discussed in Sec. III. δ < 1. Any non-signaling correlation can be reproduced with HV models
with complete OD.
Results
Relaxing the assumptions Result 1: Quantum correlations that cannot be simulated if there
Quantifying measurement dependence. To quantify any lack of MI, is arbitrarily small MI or PI
and therefore to quantify MD, we have to take into account the dis- Consider the bipartite Bell experiment in which Alice and Bob have two
tribution of x and y and, therefore, we have to consider the full dis- measurement options x, y ∈ {0, 1}, each of them with 2N possible results
tribution p(a, b, x, y) rather than only p(a, b∣x, y). The full distribution which can be expressed as a string of N bits,
p(a, b, x, y) can be reproduced with an l-measurement dependent (l- a, b ∈ {(0, 0, …, 0), (0, 0, …, 1), …, (1, 1, …, 1)}. Suppose that Alice and
MD) HV model36 if it can be reproduced with an HV model such that, Bob share the following 2N × 2N-dimensional entangled state:
for all x ∈ X, y ∈ Y, and for all λ,   
∣ψ = ∣ϕ A , B      ∣ϕ A , ð9Þ
1 1 N , BN
pðx, yjλÞ ≥ l ≥ 0: ð6Þ
where
If there are only two inputs per party, a value l = 1/4 implies that p(x, y∣λ)
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
must be uniform. Therefore, in this case, p(a, b, x, y) can be reproduced ∣ϕ = að∣01i + ∣10iÞ + 1  2a2 ∣11i, ð10Þ
with an HV model with MI in which there is no correlation between the pffiffi
51
hidden variables of the particles and the measurement settings. with a = is a two-qubit state with the first qubit in Alice’s side and
2 ,
However, this is not the case for 0≤l < 1/4. We say that p(a, b, x, y) can be the second qubit in Bob’s side. Suppose that Alice’s and Bob’s
reproduced with an HV model with partial MD if there is an (l-MD) measurements are of the form
model for some l > 0. We say that p(a, b, x, y) can only be reproduced
with an HV model with complete MD if p(a, b, x, y) cannot be repro- Aa1 , ..., aN jx = Aa1 jx      AaN jx , ð11aÞ
duced with any (l-MD) model for any l > 0; the only possible HV models
have p(x, y∣λ) = 0 for some pair of settings (x, y) and some λ. If there are
Bb1 , ..., bN jy = Bb1 jy      BbN jy , ð11bÞ
only two inputs per party, any p(a, b, x, y) corresponding to any non-

Nature Communications | (2025)16:4390 2


Article https://doi.org/10.1038/s41467-025-59247-7

where, here, the tensor product refers to the qubits in each observer’s Table 1 | Relaxation of PI as a function of the number of
system and the specific form of the factors is given by parallel copies
N ε pN
A1jx = 1  A0jx , ð12aÞ H
1 < 0.0461 0.0902
2 < 0.0901 0.1722
B1jy = 1  B0jy , ð12bÞ 3 < 0.1321 0.2469
4 < 0.1722 0.3148
where 5 < 0.2104 0.3766
6 < 0.2468 0.4328
A0j0 = B0j0 = ∣0ih0∣, ð13aÞ
7 < 0.2816 0.4839
8 < 0.3147 0.5304

A0j1 = B0j1 = ∣φ φ∣, ð13bÞ 9 < 0.3463 0.5727
10 < 0.3765 0.6113
 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
with ∣φ = pffiffiffiffiffiffiffiffi
1
ð 1  2a2 ∣0i  a∣1iÞ. That is, each of the 2N-outcome N is the number of parallel copies and 2N is the number of outputs in the Bell test. ε = εA = εB
1a2
measurements can be seen as N (nonindependent) two-outcome quantifies the relaxation of PI and pN
H is the upper bound in the probability given by Eq. (16) for l-
MD and (ε, ε)-PD HV models satisfying OI. HV models with ε above the threshold indicated in the
measurements performed simultaneously on a 2N-dimensional quan-
Table cannot be excluded by the corresponding experiment.
tum system. These state and measurements produce a correlation with
the following properties:

pð0, 1, a2 , b2 , . . . , aN , bN j0, 1Þ = . . . = pða1 , b1 , . . . , aN1 , bN1 , 0, 1j0, 1Þ = 0, the two-observer two-setting two-outcome Bell scenario39. Any other
ð14aÞ correlation with this property can also be used in the experiment. The
full characterisation of these points is in40.
A natural question is what conditions a correlation must
pð1, 0, a2 , b2 , . . . , aN , bN j1, 0Þ = . . . = pða1 , b1 , . . . , aN1 , bN1 , 1, 0j1, 0Þ = 0,
satisfy to allow for arbitrarily small MI and PI, and whether there
ð14bÞ are quantum correlations in Bell scenarios with finite number of
inputs that allow for such relaxation. In the Supplementary
pð0, 0, a2 , b2 , . . . , aN , bN j1, 1Þ = . . . = pða1 , b1 , . . . , aN1 , bN1 , 0, 0j1, 1Þ = 0, Note 3, we show a necessary condition - the quantum correlation
must necessarily lie on or be arbitrarily close to the nonsignaling
ð14cÞ
boundary. We also illustrate by an explicit example that this
for all a1, …, aN, b1, …, bN ∈ {0, 1}. Eq. (14a) indicates that, if the mea- condition is not sufficient. We leave as an open question whether
surements are x = 0 for Alice and y = 1 for Bob, then, in the N-bit strings there is a finite input-output quantum correlation that proves
that Alice and Bob obtain as outputs cannot be one position where Result 1.
Alice has 0 and Bob has 1. Similarly, for Eqs. (14b) and (14c). These state
and measurements are the ones needed for the parallelised version37 of Experimental test to exclude HV theories with partial MD and PD
the optimal version of the proof of Bell non-locality proposed by So far, we have identified a quantum correlation that cannot be
Hardy38. simulated by any l-MD and (εA, εB)-PD HV model with l > 0, εA < 1,
Let us define εB < 1, satisfying OI. This correlation is a point in the set of
X quantum correlations. The problem is that, due to experimental
 
pNH : = p a1 , b1 , . . . , aN , bN j0, 0 , errors, an actual experiment will fail to exactly produce this
a , . . . , aN , b1 , . . . , bN point. Here, we reformulate Result 1 in a way that the existence of
  1   correlations that cannot be simulated by l-MD and (εA, εB)-PD HV
a1 , b1 = ð0, 0Þ _ . . . _ aN , bN = ð0, 0Þ
models with l > 0, εA < 1, εB < 1, and satisfying OI, can be
ð15Þ
experimentally tested.

where is the logical OR. It can be proven (see Supplementary Note 4) that, for any l-MD
Result 1 can be stated as follows: In any l-MD and (εA, εB)-PD HV and (εA, εB)-PD HV model with l > 0, εA < 1, εB < 1, satisfying OI, the
model satisfying OI and Eqs. (14a), (14b) and (14c), for all l > 0 and all N, following Bell-like inequality holds:

pNH ≤ εA + εB  εA εB : ð16Þ I Nκ ðpÞ ≤ ~εA + ~εB  ~εA ~εB , ð18Þ

The proof is in the Supplementary Note 2. Therefore, if εA < 1 and εB < 1, where
then pNH < 1. In contrast, in quantum theory37, as N tends to infinity, X     
I Nκ ðpNH Þ : = p a1 , b1 , . . . , aN , bN j0, 0
N!1 a1 , . . . , aN , b1 , . . . , bN
pNH ! 1: ð17Þ    
a1 , b1 = ð0, 0Þ _ . . . _ aN , bN = ð0, 0Þ
X     
Consequently, for any l-MD and (εA, εB)-PD HV model with l > 0, εA < 1, κ p a1 , b1 , . . . , aN , bN j0, 1
a , . . . , aN , b1 , . . . , bN
εB < 1, satisfying OI, there is N such that quantum theory predicts a   1  
a1 , b1 = ð0, 1Þ _ . . . _ aN , bN = ð0, 1Þ
value for pNH that cannot be simulated. X      ð19Þ
κ p a1 , b1 , . . . , aN , bN j1, 0
For example, Table 1 gives the values of ε = εA = εB that cannot be
a , . . . , aN , b1 , . . . , bN
simulated if nature achieves the quantum value for pNH . Notice that the   1  
a1 , b1 = ð1, 0Þ _ . . . _ aN , bN = ð1, 0Þ
number of excluded HV models grows with N. As N tends to infinity, the X     
κ p a1 , b1 , . . . , aN , bN j1, 1 ,
only surviving HV models are those with ε = 1.
a1 , . . . , aN , b1 , . . . , bN
The correlations defined by Eqs. (10)–(13) are special: they define    
a1 , b1 = ð0, 0Þ _ . . . _ aN , bN = ð0, 0Þ
an extremal non-exposed point of the quantum set of correlations for

Nature Communications | (2025)16:4390 3


Article https://doi.org/10.1038/s41467-025-59247-7

with Result 2 can be formulated as follows: In any δ-OD HV model that


satisfies MI, PI and Eq. (27),
N2
κ> , ð20Þ
lð1  εÞ2 δq ð29Þ
pM
H ≤ ,
2
where ε = maxfεA , εB g, and
where q = M2+ 1. The proof is provided in the Supplementary Note 5.
rffiffiffiffiffi
2 Therefore, if δ < 1, it follows that pM 1
H < 2. In contrast, in quantum
~εA = εA + N , ð21aÞ 21,40
lκ theory , as M approaches infinity,

rffiffiffiffiffi M!1 1
2 pM
H ! : ð30Þ
~εB = εB + N : ð21bÞ 2

This means that, for any l-MD and (εA, εB)-PD HV model with l > 0, Consequently, for any HV model with MI, PI and δ-OD, with δ < 1, there
εA < 1, εB < 1, satisfying OI, for sufficiently large κ, the quantity is M such that quantum theory predicts a value for pM H that cannot be
I Nκ ðpÞ is upper bounded by a value that is always smaller than 1. simulated. In addition, it can be proven (see Supplementary Note 6)
Furthermore, for fixed N, this bound approaches the bound for that the set of correlations produced by HV with MI, PI and complete
(16) when we take large values of κ and is therefore violated by OD is the set of nonsignaling correlations.
the quantum state and measurements described earlier. The above proof is based on the assumption that Eq. (27) hold. In
an actual experiment, instead of the zeros Eq. (27), we will obtain small
Result 2: Quantum correlations which cannot be simulated if values. Once we have them, we can derive an optimal Bell-like
there is arbitrarily small OI inequality that will allows us to discard any HV model with δ-OD for
Consider the bipartite Bell experiment in which Alice and Bob have some δ < 1.
M + 1 measurement options x, y ∈ {0, 1, …, M}, each of them with 2
possible results, a, b ∈ {0, 1}. Suppose that Alice and Bob share the Discussion
following two-qubit entangled state: The results presented have consequences both for foundations and
applications in quantum information processing, communication and
 1 computation. For foundations, our results bring us closer to the
∣ϕ = pffiffiffiffiffiffiffiffiffiffiffi ðt∣00i  ∣11iÞ, ð22Þ
1 + t2 solution of a problem proposed by Shimony11[pp. 96, 124, 149]34[p. 66]
and which can be formulated as follows: “One of these three premises
where t ∈ [0, 1] is the value that maximises [MI, PI and OI] must be false and it is important to locate the false
one”11[p. 96]. If we assume that only one is false (and that it is the same
 2 one for all non-local quantum correlations), then
t2 1  t 2M ð23Þ
max : I. If the assumption that fails is MI, Result 1 shows that, MI has to fail
0 ≤ t ≤ 1 1 + t 2 1 + t 2M + 1
completely because there are quantum correlations that can only
Alice’s and Bob’s measurements are of the form Aajx = jπ ajx ihπ ajx j and be explained with complete MD.
Bbjy = jσ bjy ihσ bjy j, with II. If the assumption that fails is PI, Result 1 shows that, PI has to fail
completely because the same correlations used in [I] can only be
jπ 0jx i = cos ax j0i + sin ax j1i, 8x 2 f0, . . . , Mg, explained with complete PD.
ð24Þ III. If the assumption that fails is OI, Result 2 shows that OI has to fail
jπ 1jx i =  sin ax j0i + cos ax j1i, 8x 2 f0, . . . , Mg,
completely because there are quantum correlations that can only
and be explained with complete OD.

jσ 0jy i = cos by j0i + sin by j1i, 8y 2 f0, . . . , Mg, Each of these solutions to Shimony’s problem requires extra
ð25Þ
jσ 1jy i =  sin by j0i + cos by j1i, 8y 2 f0, . . . , Mg, causal influences which are not needed if HVs do not exist. These extra
causal influences are shown with non-black colours in Fig. 1a–c,
with respectively. The causal influences if HVs do not exist are shown
h i in Fig. 1d.
ak = bk = arctan ð1Þk t k + 1=2 8k 2 f0, . . . , Mg: ð26Þ More generally, our results allow us to experimentally narrow
down the possible explanations of Bell non-locality and the whole
These state and measurements produce a correlation with the fol- quantum theory, since they allow to experimentally excluding large
lowing properties: subsets of HV models that are not excluded by previous experiments.
Specifically, in principle, any l-MD, (εA, εB)-PD HV model with l > 0,
pð0, 0j0, 0Þ = 0, ð27aÞ εA < 1, εB < 1, satisfying OI can be experimentally excluded. Similarly,
any δ-OD HV model with δ < 1 and satisfying MI and PI can be, in
principle, experimentally excluded. Still, the experiments cannot
pð0, 1jk, k  1Þ = 0 8k 2 f1, . . . , Mg, ð27bÞ
exclude HV models with complete MD22 or complete PD26,32 or com-
plete OD.
pð1, 0jk  1, kÞ = 0 8k 2 f1, . . . , Mg, ð27cÞ Result 1 extends the observation in36 that there are quantum
correlations that cannot be obtained from an l-MD HV model that
and correspond to the optimal implementation of the “ladder” version satisfies OI and PI, for all values of l > 0. In36, the difference between
of Hardy’s proof21,38,40. quantum theory and the models with OI, PI and arbitrarily small MI is
Let us define so small that any relaxation of PI makes the difference to vanish. In this
respect, Result 1 makes testable the impossibility of HV models with
pM
H : = pð0, 0jM, MÞ: ð28Þ partial MD and PD.

Nature Communications | (2025)16:4390 4


Article https://doi.org/10.1038/s41467-025-59247-7

Fig. 1 | Space-time diagrams of the causal influences needed, in every round of cones of X and Y, represented by the upper node Λ. The causal influences between
the Bell-like test, for each of the possible solutions to Shimony’s problem in the upper Λ and X and Y are represented by violet dashed arrows. b Complete
light of our results. In all diagrams, black arrows represent causal influences parameter dependence. X is decided by Alice and Y is decided by Bob. However, X
common to all possibilities. a Complete measurement dependence. It can occur in, does not only influence Alice’s measurement outcome, represented by A, but also
essentially, two ways. The first is with complete superdeterminism without retro- Bob’s measurement outcome, represented by B, which is outside the light cone of
causality. In this case, Alice and Bob do not have freedom of choice to choose the X. This superluminal influence (or “action at a distance'') is represented by a red
measurement settings, X and Y, respectively. Instead, the settings are determined arrow. Similarly, Y does not only influence B, but also A. c Complete outcome
by the distribution of the HVs in the past light cones of X and Y, represented by the dependence. X is decided by Alice and Y is decided by Bob. However, A and B are
lower node Λ. The causal influences between the lower Λ and X and Y are repre- causally connected despite they are space-like separated. These superluminal
sented by violet continuous arrows. The second way complete MD can occur is with influences are represented by blue arrows. d No hidden variables. X is decided
complete retrocausality with freedom of choice. In this case, Alice and Bob have freely by Alice and Y is decided freely by Bob. A is causally connected only to the
freedom to choose the “nominal” measurement settings X and Y, but the actual quantum state, represented by Ψ, and X. Similarly, B is causally connected only to
measurements are determined by the distribution of the HVs in the future light Ψ and Y.

Result 2 is related to the observation in31 that there are quantum of HV models facilitates the discussion of the remaining models and, in
correlations that cannot be simulated with the assumptions of causal particular, the discussion of the thermodynamics of the HV models45
models (CM), MI, causal parameter independence (CPI) and the com- that could not be discarded.
plete relaxation of causal outcome independence (COI). This observa- Our results are also of practical interest in quantum informa-
tion is not made in the framework of the four assumptions of Bell’s tion processing, quantum communication and quantum computa-
theorem (HV, MI, PI, OI) but in the framework of causal models41. In tion. In the first place, for a general reason: the results show that
general, neither HV and CM, nor PI and CPI, nor OI and COI, are quantum correlations do not only offer advantage with respect to
equivalent. local correlations, but also with respect to correlations assisted by
In addition, Results 1 and 2 confirm that there is some inter- partial instantaneous actions at a distance or even assuming the
changeability between MD and (PD+OD)42,43. However, our results go existence of partial constraints to freedom of choice or partial
beyond that as they show that epsilon of each of MD and (PD+OD) is retrocausal influences. This can make a big difference in quantum
not enough: complete MD or complete PD or complete OD is needed. computational advantage. For example, when mapping quantum
One reason why it is important to exclude HV models with partial non-local correlations into the circuit model, the advantage of
(but not complete) MD is that these models have been proposed to quantum theory with respect to local HV theories translates into a
explain quantum correlations12,14–21. In addition, partial MD or, more non-oracular quantum advantage46,47. Our results show that there is
precisely partial human’s free will, has been proposed in philosophy to also advantage with respect to non-local correlations with partial
resolve the conflict between the concept of an omniscient God and MD, PD and OD. This may translate into new forms of quantum
God’s commandment not to commit sin44. computational advantage.
One reason why it is important to exclude HV models with partial Another reason why our results are of practical interest is device-
(but not complete) actions at a distance is that these models have been independent (DI) quantum information processing48,49. DI protocols
proposed to explain quantum correlations27–32. A second reason, which for random number generation50, quantum key distribution48, state
also applies to HV models with partial MD, is that excluding larger sets tomography51 and self-testing of quantum devices52 achieve advantage

Nature Communications | (2025)16:4390 5


Article https://doi.org/10.1038/s41467-025-59247-7

allowing users to monitor the performance of their devices irrespec- 16. Brandão, F. G. S. L. et al. Realistic noise-tolerant randomness
tive of noise, imperfections, and lack of knowledge regarding the inner amplification using finite number of devices. Nat. Commun. 7,
workings - the users simply treat their devices as black boxes with 11345 (2016).
classical inputs and outputs. An obstacle for practical DI protocols is 17. Hall, M. J. W. Complementary contributions of indeterminism and
the experimentally challenging requirement of a Bell test with: (I) signaling to quantum correlations. Phys. Rev. A 82, 062117 (2010).
quantum devices being isolated from each other, (II) with the inputs 18. Barrett, J. & Gisin, N. How much measurement independence is
being chosen with uniform randomness and (III) with the detection needed to demonstrate nonlocality? Phys. Rev. Lett. 106,
loophole53 closed. Experiments in different platforms54–56 allow for Bell 100406 (2011).
tests with the detection loophole closed and high DI randomness 19. Ramanathan, R. et al. Randomness amplification under minimal
generation rates. However, in these platforms, the quantum systems fundamental assumptions on the devices. Phys. Rev. Lett. 117,
are very close, primarily to drive high entanglement generation rates 230501 (2016).
via non-negligible coupling. The problem is that, precisely because the 20. Hall, M. J. W. & Branciard, C. Measurement-dependence cost for
systems are close to one another, they can no longer be regarded as Bell nonlocality: Causal versus retrocausal models. Phys. Rev. A
isolated in the sense needed for a Bell test. Sophisticated theoretical 102, 052228 (2020).
techniques have been devised to handle the issues of cross-talk and 21. Ramanathan, R., Banacki, M. & Horodecki, P. No-signaling-proof
weak seeds separately. A Bell-like test allowing for arbitrary relaxation randomness extraction from public weak sources. arXiv preprint
of MI and PI provides a simple and elegant solution to the problem of arXiv:2108.08819 (2021).
leakage of input information. A Bell-like test allowing for simultaneous 22. Brans, C. Bell’s theorem does not eliminate fully causal hidden
relaxation of MI, PI and OI would allow DI randomness generation variables. Int. J. Theor. Phys. 27, 219–226 (1988).
tolerating weak seeds and cross-talk. We hope that our results will 23. ’t Hooft, G. The Cellular Automaton Interpretation of Quantum
stimulate research in these directions. Mechanics, vol. 185 298 of Fundamental Theories of Physics
(Springer Cham, 2016).
Data availability 24. Costa de Beauregard, O. A response to the argument directed by
No data sets were generated or analysed during the current study. einstein, poldosky and rosen against the bohrian interpretation of
quantum phenomena. C. R. Acad. Sci. 236, 1632–1634 (1953).
References 25. Donadi, S. & Hossenfelder, S. Toy model for local and deterministic
1. Born, M. Quantenmechanik der stoßvorgänge. Z. Physik 38, wave-function collapse. Phys. Rev. A 106, 022212 (2022).
803–827 (1926). 26. Bohm, D. A suggested interpretation of the quantum theory in terms
2. Bell, J. S. On the Einstein Podolsky Rosen paradox. Physics 1, of “hidden” variables. I. Phys. Rev. 85, 166–179 (1952).
195–200 (1964). 27. Brassard, G., Cleve, R. & Tapp, A. Cost of exactly simulating quan-
3. Clauser, J. F., Horne, M. A., Shimony, A. & Holt, R. A. Proposed tum entanglement with classical communication. Phys. Rev. Lett.
experiment to test local hidden-variable theories. Phys. Rev. Lett. 83, 1874 (1999).
23, 880–884 (1969). 28. Bacon, D. & Toner, B. F. Bell inequalities with auxiliary commu-
4. Einstein, A., Podolsky, B. & Rosen, N. Can quantum-mechanical nication. Phys. Rev. Lett. 90, 157904 (2003).
description of physical reality be considered complete? Phys. Rev. 29. Pironio, S. Violations of Bell inequalities as lower bounds on the
47, 777–780 (1935). communication cost of nonlocal correlations. Phys. Rev. A 68,
5. Freedman, S. J. & Clauser, J. F. Experimental test of local hidden- 062102 (2003).
variable theories. Phys. Rev. Lett. 28, 938–941 (1972). 30. Pawłowski, M., Kofler, J., Paterek, T., Seevinck, M. & Brukner, Č. Non-
6. Aspect, A., Dalibard, J. & Roger, G. Experimental test of Bell’s local setting and outcome information for violation of Bell’s
inequalities using time-varying analyzers. Phys. Rev. Lett. 49, inequality. New J. Phys. 12, 083051 (2010).
1804–1807 (1982). 31. Ringbauer, M. et al. Experimental test of nonlocal causality. Sci.
7. Hensen, B. et al. Loophole-free Bell inequality violation using elec- Adv. 2, e1600162 (2016).
tron spins separated by 1.3 kilometres. Nature 526, 682–686 (2015). 32. Brask, J. B. & Chaves, R. Bell scenarios with communication. J. Phys.
8. Giustina, M. et al. Significant-loophole-free test of bell’s theorem A: Math. Theor. 50, 094001 (2017).
with entangled photons. Phys. Rev. Lett. 115, 250401 (2015). 33. Blasiak, P. Classical systems can be contextual too: Analogue of the
9. Shalm, L. K. et al. Strong loophole-free test of local realism. Phys. mermin–peres square. Ann. Phys. 353, 326–339 (2015).
Rev. Lett. 115, 250402 (2015). 34. Shimony, A.Search for a Naturalistic World View. Volume I: Scientific
10. Jarrett, J. P. On the physical significance of the locality conditions in Method and Epistemology (Cambridge University Press, Cambridge,
the Bell arguments. Noûs 18, 569–589 (1984). UK, 1993).
11. Shimony, A. Search for a Naturalistic World View. Volume II: Natural 35. Polkinghorne, J. Physics and theology. Europhys. News 45 (2014).
Science and Methaphysics (Cambridge University Press, Cam- 36. Pütz, G., Rosset, D., Barnea, T. J., Liang, Y.-C. & Gisin, N. Arbitrarily
bridge, UK, 1993). small amount of measurement independence is sufficient to man-
12. Hall, M. J. W. The significance of measurement independence for ifest quantum nonlocality. Phys. Rev. Lett. 113, 190402 (2014).
Bell inequalities and locality. In Asselmeyer-Maluga, T. (ed.) At the 37. Mančinska, L. & Vidick, T. Unbounded entanglement in nonlocal
Frontier of Spacetime: Scalar-Tensor Theory, Bells Inequality, games. Quantum Inf. Comput. 15, 1317–1332 (2015).
Machs Principle, Exotic Smoothness, vol. 183 of Fundamental The- 38. Hardy, L. Nonlocality for two particles without inequalities for
ories of Physics, 189-204 (Springer, Switzerland, 2016). almost all entangled states. Phys. Rev. Lett. 71, 1665–1668
13. Bell, J. S., Shimony, A., Horne, M. A. & Clauser, J. F. An exchange on (1993).
local beables. Dialectica 39, 85–110 (1985). 39. Goh, K. T. et al. Geometry of the set of quantum correlations. Phys.
14. Hall, M. J. W. Relaxed Bell inequalities and Kochen-Specker theo- Rev. A 97, 022104 (2018).
rems. Phys. Rev. A 84, 022102 (2011). 40. Zhao, S., Ramanathan, R., Liu, Y. & Horodecki, P. Tilted Hardy
15. Colbeck, R. & Renner, R. Free randomness can be amplified. Nat. paradoxes for device-independent randomness extraction. Quan-
Phys. 8, 450–453 (2012). tum 7, 1114 (2023).

Nature Communications | (2025)16:4390 6


Article https://doi.org/10.1038/s41467-025-59247-7

41. Wood, C. J. & Spekkens, R. W. The lesson of causal discovery “New tools in quantum information and communication” (Project No.
algorithms for quantum correlations: causal explanations of Bell- PID2020-113738GB-I00), and the Digital Horizon Europe project "Foun-
inequality violations require fine-tuning. New J. Phys. 17, dations of quantum computational advantage” (FoQaCiA) (Grant
033002 (2015). agreement No. 101070558).
42. Conway, J. H. & Kochen, S.The strong free will theorem, 443–454
(Cambridge University Press, 2011). Author contributions
43. Blasiak, P., Pothos, E. M., Yearsley, J. M., Gallus, C. & Borsuk, E. C.V., R.R., and A.C. contributed equally to this work.
Violations of locality and free choice are equivalent resources in
Bell experiments. Proc. Natl. Acad. Sci. USA 118, Competing interests
e2020569118 (2021). There are no competing interests.
44. Specker, E. Die Logik nicht gleichzeitig entscheidbarer Aussagen.
Dialectica 14, 239–246 (1960). Additional information
45. Cabello, A., Gu, M., Gühne, O., Larsson, J.-A. & Wiesner, K. Ther- Supplementary information The online version contains
modynamical cost of some interpretations of quantum theory. Phys. supplementary material available at
Rev. A 94, 052127 (2016). https://doi.org/10.1038/s41467-025-59247-7.
46. Bravyi, S., Gosset, D. & König, R. Quantum advantage with shallow
circuits. Science 362, 308–311 (2018). Correspondence and requests for materials should be addressed to
47. Bravyi, S., Gosset, D., König, R. & Tomamichel, M. Quantum Carlos Vieira or Ravishankar Ramanathan.
advantage with noisy shallow circuits. Nat. Phys. 16,
1040–1045 (2020). Peer review information Nature Communications thanks the anon-
48. Liu, Y. et al. Device-independent quantum random-number gen- ymous reviewer(s) for their contribution to the peer review of this work. A
eration. Nature 562, 548–551 (2018). peer review file is available
49. Zhang, W. et al. A device-independent quantum key distribution
system for distant users. Nature 607, 687–691 (2022). Reprints and permissions information is available at
50. Pironio, S. et al. Random numbers certified by Bell’s theorem. Nat- http://www.nature.com/reprints
ure 464, 1021–1024 (2010).
51. Mayers, D. & Yao, A. Self testing quantum apparatus. Quantum Info. Publisher’s note Springer Nature remains neutral with regard to jur-
Comput. 4, 273–286 (2004). isdictional claims in published maps and institutional affiliations.
52. Magniez, F., Mayers, D., Mosca, M. & Ollivier, H. Self-testing of
quantum circuits. In Bugliesi, M., Preneel, B., Sassone, V. & Wege- Open Access This article is licensed under a Creative Commons
ner, I. (eds.) Automata, Languages and Programming, 72–83 Attribution-NonCommercial-NoDerivatives 4.0 International License,
(Springer Berlin Heidelberg, Berlin, Heidelberg, 2006). which permits any non-commercial use, sharing, distribution and
53. Pearle, P. M. Hidden-variable example based upon data rejection. reproduction in any medium or format, as long as you give appropriate
Phys. Rev. D 2, 1418–1425 (1970). credit to the original author(s) and the source, provide a link to the
54. Rowe, M. A. et al. Experimental violation of a Bell’s inequality with Creative Commons licence, and indicate if you modified the licensed
efficient detection. Nature 409, 791–794 (2001). material. You do not have permission under this licence to share adapted
55. Ansmann, M. et al. Violation of Bell’s inequality in Josephson phase material derived from this article or parts of it. The images or other third
qubits. Nature 461, 504–506 (2009). party material in this article are included in the article’s Creative
56. Arute, F. et al. Quantum supremacy using a programmable super- Commons licence, unless indicated otherwise in a credit line to the
conducting processor. Nature 574, 505–510 (2019). material. If material is not included in the article’s Creative Commons
licence and your intended use is not permitted by statutory regulation or
Acknowledgements exceeds the permitted use, you will need to obtain permission directly
We thank Paweł Horodecki and Pedro Lauand for discussions and from the copyright holder. To view a copy of this licence, visit http://
acknowledge support from the Early Career Scheme (ECS) (Grant No. creativecommons.org/licenses/by-nc-nd/4.0/.
27210620), the General Research Fund (GRF) (Grant No. 17211122), the
Research Impact Fund (RIF) Grant No. R7035-21), the MCINN/AEI project © The Author(s) 2025

Nature Communications | (2025)16:4390 7

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy