betaFeedback

WorksPhysics and Astronomy

Bell scenarios encode classical non-causality

Abstract

Bell and contextuality scenarios are unified within the Acín-Fritz-Leverrier-Sainz hypergraph formalism, in which measurements are hyperedges. Multipartite measurements are obtained from product rules that combine independent contextuality scenarios into a single hypergraph. Several such rules exist: they are structurally distinct, yet yield the same classical, quantum, and boxworld correlations. Here, we show that this difference is not incidental, unveiling an unexpected connection. The measurements of composite Bell scenarios are exactly the process functions, the deterministic classical communication protocols that may be non-causal, i.e., lack a definite causal order without generating paradoxes. Validity of a hyperedge and logical consistency of a process function are thus one and the same condition. This allows us to characterize process functions by normalization and no-signaling alone, providing transformations that generate non-causal process functions from no-communication ones. The product rules, in turn, order process functions by causal structure, from no-communication to non-causality, which yields a graphical criterion for the latter: a process function is non-causal precisely when its hyperedge is generated by the disjunctive but not the maximal product. We thereby obtain the first joint hypergraph representation of (non-causal) process functions and (nonlocal-without-entanglement) measurements, recasting classical non-causality as a non-standard way of composing black boxes.

Introduction

Quantum theory allows for statistics that admit no classical explanation. Such non-classicality famously manifests itself as Bell nonlocality [1, 2, 3] and Kochen–Specker (KS) contextuality [4, 5]. Bell nonlocality is certified when the correlations of spatially separated parties, though respecting no-signaling, cannot be reproduced by any local causal model. The KS theorem is a closely related but originally algebraic result, establishing the impossibility of constructing a deterministic assignment of outcomes to quantum measurements that is compatible with their algebraic properties. These two results are brought together within various frameworks, including ontological [6], graph-theoretic [7], and sheaf-theoretic [8] ones. The latter two are unified in the Acín-Fritz-Leverrier-Sainz (AFLS) formalism [9] that provides a theory-independent, hypergraph-based approach to contextuality.

The AFLS framework reformulates contextuality operationally: collections of measurements are described by hypergraphs, in which vertices represent events (measurement outcomes) and hyperedges represent measurements. Such a scenario is also called a test space [10, 11]. Contextuality is not a property of the scenario itself, but of the empirical statistics defined on it: the latter are called contextual if they cannot be written as a convex combination of deterministic {0,1}\{0,1\}-valued assignments to events, even though the no-disturbance principle [12, 13] is satisfied, i.e. the statistics of a measurement does not depend on which compatible measurements are performed alongside it. In this language, the no-signaling principle and Bell nonlocality are the special cases of no-disturbance and Kochen-Specker contextuality, respectively: the context is fixed by the choices of spatially separated parties, i.e, multipartite Bell scenarios are built by composing local ones.

This raises a natural question: given several independent contextuality scenarios, how should one compose them into a single hypergraph representing their joint scenario? The composition is non-trivial and should require that probabilistic models on the composite scenario respect the no-signaling principle. For two scenarios, Foulis and Randall [14] identified the unique composition rule with this property. For more than two scenarios, however, several rules satisfy no-signaling [9, 15], forming a family of Foulis–Randall (FR) products. Despite their formal differences, these products are observationally equivalent at the level of classical, quantum, and boxworld [16, 17] correlations, i.e., the respective sets of correlations do not change under the composition rules [15]. The one known exception concerns almost-quantum correlations [18], which coincide across the common, maximal, and disjunctive FR products but differ for the minimal one [15]. The minimal, common, and maximal products admit an operational reading: the respective hyperedges correspond to measurement protocols, i.e., to a specific order in which the parties choose measurements and communicate their outcomes to one another. For the disjunctive FR product, however, some hyperedges cannot be interpreted as standard measurement protocols. Whether these products carry any further physical significance has not been addressed so far.

As shown here, the answer comes from an a priori independent research line: the study of indefinite causal order [19]. This phenomenon is investigated within the process matrix formalism [20]: each party performs ordinary quantum operations in its local laboratory, while these operations are linked by a global process that need not admit any consistent ordering of the parties. Some processes even violate causal inequalities [20, 21], a feature known as non-causality. More specifically, what matters here are process functions [22, 23, 24, 25, 26], the deterministic classical analogues of process matrices. Process functions model all single-round, deterministic classical processes among the parties that are logically consistent, in the sense of not generating time-travel paradoxes while preserving each party’s freedom of choice [27]. Strikingly, there exist process functions that are non-causal, that is they remain consistent even if they do not admit any consistent ordering of the parties. Several families of non-causal process functions for n≥3n \ge3 parties are known [22, 25, 26, 28, 30, 31] [29], the archetype being the Lugano (AF/BW) process [23] [29].

Results

In this work, we uncover a sharp operational distinction between the different FR products, rooted in causality: each measurement they generate defines a process function, with each FR product singling out a different class of causal structures, which form a strict hierarchy.

After a general presentation of the AFLS formalism (Section III), we begin, for ease of exposition, with the bipartite case, where we establish a one-to-one correspondence between the hyperedges generated by the original FR product (defined in Section IV A) and bipartite process functions, i.e. no-communication and one-way-communication processes (Section IV B, Proposition 1). We then turn to the general case. Relying on a recently introduced combinatorial characterization of process functions [32] as complete pairwise exclusive sets of multipartite events, we show that process functions and the hyperedges generated by the disjunctive FR product [15] of local unipartite Bell scenarios are one and the same objects (Section V, Theorem 4).

We then explore three implications of this equivalence (Section VI).

First, it implies that the validity criteria of process functions and hyperedges of the disjunctive FR product coincide (Section VI A, Corollary 1). In particular, it recasts pairwise exclusive completeness [32] as a characterization of process functions in terms of normalization and no-signaling conditions (Section VI A, Corollary 2). Second, it sheds new light on the hierarchy of composite contextuality scenarios generated by the different FR products [9, 15] (Section VI B), which turns out to mirror a hierarchy of causal structures. The minimal product corresponds to space-like separated parties sharing a common future; the common product additionally includes all fixed causal orders; the maximal product further encompasses all causal process functions, including deterministic classical control of causal order; and the disjunctive product finally admits non-causal structures. This yields a graphical criterion for non-causality: a process function is non-causal precisely when the corresponding hyperedge is generated by the disjunctive but not the maximal FR product (Corollary 3). Moreover, we relate this hierarchy to that of quantum circuits with quantum control of causal order (QC-QC) [33].

Third, inspired by previous work [15, 34], we indicate how the normalization and no-signaling conditions characterizing process functions yield no-signaling substitution rules which allow to generate non-causal process functions from no-communication ones (Section VI C). As an example, starting from a minimal hyperedge associated with a no-communication process function, we derive the hyperedge associated with the Lugano process.

Taken together, our results provide the first joint hypergraph representation of classical non-causality and nonlocality without entanglement – two phenomena whose equivalence was established in the case of quantum measurements in [26, 35]. They also reveal that classical non-causality can be understood – in line with previous work [34] – as a non-standard way of composing no-signaling measurements, complementing its usual reading as the classical counterpart of quantum indefinite causal order [19, 20].

The AFLS hypergraph framework

Following [9], we first recall the AFLS framework. The AFLS framework relies on the notion of hypergraphs.

Definition 1 (Hypergraph). A hypergraph is a pair H=(V(H),E(H))H = (V(H), E(H)), where V(H)V(H) is a finite set of vertices, and E(H)⊆2V(H)∖∅E(H) \subseteq2^{V(H)} \setminus\varnothing is a collection of non-empty subsets of VV, called hyperedges.

Unlike ordinary graphs, whose edges connect exactly two vertices, the hyperedges defining a hypergraph may contain an arbitrary number of vertices in general. Each hyperedge then represents a measurement, i.e., a set of mutually connected vertices corresponding to its mutually exclusive measurement outcomes (cf. Figure 1a, top). The hyperedges may share outcomes (cf. Figure 1b): the same outcome can appear in several hyperedges, each corresponding to a different measurement context in which it is obtained. A contextuality scenario is then defined as a specific type of hypergraph:

Figure 1. The AFLS hypergraph framework for a single party. (a) A contextuality scenario is described by a hypergraph, defined by hyperedges that connect a set of vertices. Here, a single hyperedge ee (blue box) is shown, which contains four vertices viv_i, i=1,…,4i=1,\ldots,4. ee is interpreted as a measurement and the vertices as its outcomes. A probabilistic model (lower part of (a)) assigns a probability to each vertex such that normalization is satisfied, as indicated by the bars over the vertices and the red legend. This defines a measurement in a theory-independent way. (b) shows two examples of contextuality scenarios that are contained in the AFLS approach: outcomes are shared across hyperedges. The pentagon scenario (bottom) also allows for virtual hyperedges (orange dashed line), which are implied by the normalization conditions (cf. Eq. (3) and Appendix A). Although virtual hyperedges are not required for the main results of this work, they need to be considered for general contextuality scenarios. Additional details are provided in Appendix A. (c) A unipartite Bell scenario B1,m,dB_{1,m,d} consists of mm disjoint hyperedges exe_x, with x∈{0,…,m−1}x \in\{0,\ldots,m-1\}, corresponding to mm measurement settings, each containing dd vertices or rather measurement outcomes. Each vertex is labeled by the corresponding unipartite event (a∣x)(a|x), specifying the outcome a∈[d]a \in[d] obtained upon choosing the measurement setting x∈[m]x \in[m], where [k]={0,…,k−1}[k] = \{0,\ldots,k-1\} (cf. Figure 1c).

Definition 2 (Contextuality scenario). A contextuality scenario is a hypergraph H=(V(H),E(H))H = (V(H), E(H)) such that

⋃e∈E(H)e=V(H).(1)\bigcup_{e \in E(H)} e = V(H). \tag*{(1)}

The condition in Eq. (1) ensures that every outcome belongs to at least one measurement. Contextuality scenarios coincide with the notions of test spaces [10, 11] and related structures in quantum logic [36].

Measurement statistics are described by assigning a probability to each outcome. This is formalized in the following definition.

Definition 3 (Probabilistic model). A probabilistic model on a contextuality scenario HH is a map p:(V(H),E(H))→[0,1]p : (V(H), E(H)) \to[0,1] satisfying the following conditions:

(i) No-disturbance principle. Every outcome is given a unique probability assignment, i.e. for any event vv shared by two measurements e,e′∈E(H)e,e' \in E(H),

p(v∣e)=p(v∣e′)=p(v)∀v∈e∩e′,(2)p(v|e) = p(v|e') = p(v) \quad\forall v \in e \cap e', \tag*{(2)}

such that p:(V(H),E(H))→[0,1]p : (V(H), E(H)) \to[0,1] can be simplified as p:V(H)→[0,1]p : V(H) \to[0,1].

(ii) Normalization. Every measurement e∈E(H)e \in E(H) is normalized,

∑v∈ep(v)=1∀e∈E(H).(3)\sum_{v \in e} p(v) = 1 \quad\forall e \in E(H). \tag*{(3)}

In (Figure 1a, bottom) we provide a probabilistic model satisfying the normalization condition (3). Note that condition (2) is implied by the original definitions provided in [9], while it is not explicitly mentioned. In a scenario with measurement contexts, i.e. where different measurements share outcomes and thus hyperedges overlap (cf. Figure 1b, top), it ensures that each vertex is assigned a unique probability. The set of all probabilistic models on HH is denoted by G(H)\mathcal{G}(H).

Notably, the AFLS formalism is independent of the theory that defines G(H)\mathcal{G}(H): different physical theories simply correspond to different classes of probabilistic models on the same scenario. Classical models, for instance, are convex combinations of deterministic assignments, while quantum models arise from projective measurements on a Hilbert space. Kochen-Specker contextuality is then naturally represented: the Kochen-Specker theorem [4] states that some scenarios with overlapping hyperedges admit a quantum probabilistic model but no classical ones [5, 9].

In this work, we focus on Bell scenarios: multipartite no-signaling scenarios where each local party has access to a certain number of independent local measurements. A unipartite Bell scenario, denoted by B1,m,dB_{1,m,d}, is defined by mm measurement settings, each with dd outcomes, with distinct measurements (or rather local operations) sharing no outcomes. Therefore, as a hypergraph, B1,m,dB_{1,m,d} consists of mm disjoint (i.e. non-overlapping) hyperedges, each containing dd vertices. We label each vertex by the corresponding unipartite event (a∣x)(a|x), specifying the outcome a∈[d]a \in[d] obtained upon choosing the measurement setting x∈[m]x \in[m], where [k]={0,…,k−1}[k] = \{0,\ldots,k-1\} (cf. Figure 1c). Note that for the results of this work, it suffices to restrict to the simpler case of each measurement having the same number of outcomes. The more general case of setting-dependent number of outcomes is naturally contained in the AFLS picture, by simply enlarging the number of vertices for each hyperedge.

The Foulis-Randall product generates all bipartite process functions

We first recall the definition and representation of bipartite Bell scenarios within the AFLS framework in Section IV A, and then demonstrate the equivalence of their hyperedges with bipartite process functions in Section IV B.

Bipartite Bell scenarios

To go beyond the unipartite case, one must specify how to compose local contextuality scenarios. Given two scenarios HA,HBH_A,H_B, a composition rule ⊗\otimes determines a composite scenario HA⊗HBH_A \otimes H_B. The vertices of the composite scenario are naturally identified with pairs of local vertices,

V(HA⊠HB)=V(HA)×V(HB)(4)V(H_A \boxtimes H_B)=V(H_A)\times V(H_B) \tag*{(4)}

while the choice of hyperedges depends on the particular composition rule. A probabilistic model on the composite scenario must satisfy normalization and the no-disturbance principle of Definition 3. In the composite scenario, the latter becomes the no-signaling principle [13]: marginalizing over one party’s measurement must yield a well-defined outcome probability for the other parties, i.e. the marginal statistics of one party do not depend on the other party’s measurement choice:

∑w∈ep(v,w)=∑w∈e′p(v,w)∀v∈V(HA), e,e′∈E(HB)(5)\sum_{w\in e} p(v,w) = \sum_{w\in e'} p(v,w) \qquad\forall v\in V(H_A),\ e,e'\in E(H_B) \tag*{(5)}
∑v∈ep(v,w)=∑v∈e′p(v,w)∀w∈V(HB), e,e′∈E(HA).(6)\sum_{v\in e} p(v,w) = \sum_{v\in e'} p(v,w) \qquad\forall w\in V(H_B),\ e,e'\in E(H_A). \tag*{(6)}

For unipartite Bell scenarios HA=B1,mA,dAH_A = B_{1,m_A,d_A} and HB=B1,mB,dBH_B = B_{1,m_B,d_B}, whose vertices are labeled by local events (a∣x)(a|x) and (b∣y)(b|y), respectively, a vertex of the composite scenario is labeled by a pair of local events,

((a∣x),(b∣y))=:(ab∣xy),(7)((a|x),(b|y)) =: (ab|xy), \tag*{(7)}

i.e., a bipartite event representing the joint occurrence of outcomes a,ba,b for local measurement settings x,yx,y.

As shown in Ref. [9], the naive composition choice, the Cartesian product HA×HBH_A \times H_B, is insufficient – it admits probabilistic models that violate no-signaling (Figure 2). The reason is that the Cartesian products only encodes simultaneous, non-communicating measurements (cf. Figure 3a). Enforcing no-signaling requires enlarging the hyperedge set to include all measurements with one-way no-communication – including the ones where one party’s setting may depend on the other’s outcome [14]. For instance, Figure 3b shows, in brown, one such hyperedge: with x=0x = 0 fixed, BB’s setting yy depends on AA’s outcome aa, giving the events {(00∣00),(01∣00),(10∣01),(11∣01)}\{(00|00),(01|00),(10|01),(11|01)\}. Including this hyperedge rules out the probabilistic model of Figure 2, since it assigns that hyperedge a total probability of 2, violating normalization. This is precisely the content of the Foulis–Randall product [9, 14]:

Figure 2. The Cartesian product of two local contextuality scenarios leads to signaling. Two unipartite Bell scenarios HAH_A and HBH_B, defined by two measurements settings each having two outcomes, are composed via the Cartesian product HA×HB=(VA×VB,E(HA)×E(HB))H_A \times H_B = (V_A \times V_B, E(H_A) \times E(H_B)). One can define a deterministic probabilistic model compatible with the global hypergraph (the probabilities assigned to the vertices are indicated in light red) such that the marginal outcome probabilities for AA depend on the measurement setting of BB. This is the case, for instance, in the first row for outcome 0 and setting 0 for AA.

Figure 3. Recovering the CHSH scenario from the bipartite Foulis-Randall (FR) product. (a) Two unipartite Bell scenarios (cf. Figure 1c) HAH_A for party AA and HBH_B for party BB are depicted as rounded rectangles on the left and top, with the respective local hyperedges (labeled by the measurement setting) colored in blue (HAH_A) and magenta (HBH_B). The composite vertex set is computed via the Cartesian product. A composite, no-communication hyperedge is depicted as a black square (marked by A↛BA \not\to B, B↛AB \not\to A), where both AA and BB choose the measurement settings zero. (b) shows hyperedges where AA signals to BB (A→BA \to B, light-brown solid line) and BB signals to AA (B→AB \to A, violet dashed line). (c) Depiction of the complete CHSH scenario, showing all hyperedges that are compatible with normalization and no-signaling, encompassing no-communication (black) and one-way communication (violet, light-brown).

Definition 4 (Foulis–Randall product). The FR product HA⊗FRHBH_A \mathop{\otimes}^{FR} H_B of two contextuality scenarios HAH_A and HBH_B has vertex set

V(HA⊗FRHB)=V(HA)×V(HB)(8)V(H_A \mathop{\otimes}^{FR} H_B) = V(H_A) \times V(H_B) \tag*{(8)}

and edge set E(HA⊗FRHB)=EB↛A∪EA↛BE(H_A \mathop{\otimes}^{FR} H_B) = E_{B\not\to A} \cup E_{A\not\to B}, where

EB↛A:={⋃v∈eA{v}×f(v):eA∈EA, f:eA→EB},(9)E_{B\not\to A} := \left\{ \bigcup_{v\in e_A} \{v\} \times f(v) : e_A \in E_A,\ f:e_A\to E_B \right\}, \tag*{(9)}
EA↛B:={⋃v∈eBf(v)×{v}:eB∈EB, f:eB→EA}.(10)E_{A\not\to B} := \left\{ \bigcup_{v\in e_B} f(v) \times\{v\} : e_B \in E_B,\ f:e_B\to E_A \right\}. \tag*{(10)}

Here EB↛AE_{B\not\to A} collects the measurements in which BB cannot signal to AA: the setting of BB may depend on the outcome of AA (non-constant ff), but not conversely. This includes both the simultaneous measurements eA×eBe_A \times e_B (constant ff, no communication) and those with genuine communication from AA to BB. Symmetrically, EA↛BE_{A\not\to B} collects the measurements in which AA cannot signal to BB. The intersection of both edge sets are the simultaneous measurements eA×eBe_A \times e_B of the Cartesian product.

Applying the Foulis-Randall composition rule to two unipartite Bell scenarios with two inputs and two outputs yields the CHSH scenario [37] (Figure 3c; see also Figure 7f of Ref. [9]):

Figure 7. A hierarchy of process functions from multipartite Foulis-Randall (FR) products. The nn-partite composite contextuality scenarios are constructed from the FR products of nn unipartite Bell scenarios. The process functions defined via the minimal, common, and maximal FR product are causal. The disjunctive FR product also contains hyperedges that correspond to non-causal process functions, e.g., the three-partite Lugano process (cf. Eq. (35)). The innermost layer is defined by the minimal FR product, corresponding to “all-to-one” process functions, where all parties are communicating to one party. The underlying causal structure is depicted by a causal graph (nodes are the parties, directed edges are the causal connections or rather, the functional dependencies between inputs and outputs of the corresponding process function). The next level contains process functions defined by the common FR product. These show a type of “causal memory” in the sense that the causal order is predefined and each party chooses a measurement based on the outcomes of the parties that precede in the measurement protocol. Next, the maximal FR product encompasses all causal process functions, while allowing the possibility of one party controlling the order of the other parties based on its measurement outcome (“dynamical control” of causal order). Finally, the disjunctive FR product includes also non-causal process functions, i.e., the possibility of two-way communication between all parties. By determining if a given process function is contained in the disjunctive but not the maximal FR product, one can decide if a process function is causal or not.

B2,2,2=B1,2,2⊗FRB1,2,2.(11)B_{2,2,2} = B_{1,2,2} \mathop{\otimes}^{FR} B_{1,2,2}. \tag*{(11)}

The probabilistic models over HA⊗FRHBH_A \mathop{\otimes}^{FR} H_B are then exactly the bipartite no-signaling models, i.e. the states of boxworld (Proposition 3.1.4 in [9] and Corollary 3.5 in [38]), a result that extends to the multipartite setting (Corollary 3.8 in [9]).

Note that imposing one-way no-communication at the level of hypergraphs and imposing causal order via no-signaling constraints leads to equivalent constraints on probabilistic models. Consider, for instance, the hyperedge set of Eq. (9), which allows communication from AA to BB. The normalization condition for these hyperedges can be shown to be equivalent to the conditions that impose no-signaling from BB to AA. This imposes the causal order A≺BA \prec B (i.e., AA is in the causal past of BB). We refer to [9, 38], and section 7.2.2 of [39] for more details.

Equivalence of bipartite process functions and Foulis-Randall hyperedges

While Foulis and Randall showed that one-way communication is what allows a bipartite scenario to satisfy no-signaling – a fact naturally represented within the AFLS framework – the connection between the latter and process functions has not been made before.

Consider a classical communication scenario as depicted in Figure 4, where two parties labeled AA and BB are embedded in an environment or rather, a shared communication resource. They receive an input system from the environment, described by discrete variables x∈Xx \in\mathcal{X} for party AA and y∈Yy \in\mathcal{Y} for party BB, and send back an output variables a∈Aa \in\mathcal{A} and b∈Bb \in\mathcal{B}, respectively. They also receive external inputs iA∈IAi_A \in\mathcal{I}_A, iB∈IBi_B \in\mathcal{I}_B and produce outcomes oA∈OAo_A \in\mathcal{O}_A, oB∈OBo_B \in\mathcal{O}_B by intervening locally on the input system. Note that our notation differs from previous works on process functions [22, 24, 25, 35, 40]. For each party, the intervention may depend both on the input from the environment (xx or yy) and the external input (iAi_A or iBi_B). This is described by the local stochastic channels P(a,oA∣x,iA)P(a,o_A|x,i_A) for party AA and P(b,oB∣y,iB)P(b,o_B|y,i_B) for party BB. Notably, we restrict to single-round communication, i.e., each party receives, sends and acts on a system only once. The observed correlations between </final> questions and answers are given by

Sketch of the bipartite process-function communication scenario

Figure 4. Sketch of the bipartite process-function communication scenario. Two isolated parties implement arbitrary local interventions fAf_A and fBf_B. The single-round causal structure between the parties is specified by a process function ww, represented by a classical channel ww together with retro-causal identity channels. The process function provides the incoming variables xx and yy to parties AA and BB, respectively, while the parties also receive external inputs iAi_A and iBi_B. For fixed local interventions, the parties generate external outputs oAo_A and oBo_B and output variables aa and bb, which are returned to the process function ww. In this work, following Refs. [24, 25, 26], we consider trivial external inputs and outputs, which can thus be neglected.

P(oA,oB∣iA,iB)=∑a,b,x,yP(a,oA∣x,iA)P(b,oB∣y,iB)×P(x,y∣a,b),(12)P(o_A,o_B|i_A,i_B)=\sum_{a,b,x,y}P(a,o_A|x,i_A)P(b,o_B|y,i_B) \times P(x,y|a,b), \tag*{(12)}

where P(x,y∣a,b)P(x,y|a,b) denotes the classical process that is carried out by the environment. We restrict our attention to process functions: classical shared communication environments that are logically consistent, i.e. never generate paradoxes under arbitrary local operations [27].

Definition 5 (Bipartite process function). A bipartite process function is a pair of local functions w:=(wA,wB):A×B→X×Y\mathbf{w}:=(w_A,w_B):\mathcal{A}\times\mathcal{B}\to\mathcal{X}\times\mathcal{Y}, corresponding to a deterministic classical process P(x,y∣a,b)=δ(x,y),w(a,b)P(x,y|a,b)=\delta_{(x,y),w(a,b)}, that is logically consistent: for every choice of local operations, P(a,oA∣x,iA)P(a,o_A|x,i_A) and P(b,oB∣y,iB)P(b,o_B|y,i_B), Eq. (12) yields a valid conditional probability distribution P(oA,oB∣iA,iB)P(o_A,o_B|i_A,i_B) (i.e. P(oA,oB∣iA,iB)≥0P(o_A,o_B|i_A,i_B)\ge0 and ∑oA,oBP(oA,oB∣iA,iB)=1\sum_{o_A,o_B}P(o_A,o_B|i_A,i_B)=1 for all iA,iBi_A,i_B).

The process function scenario depicted in Figure 4 also involves deterministic local interventions. Following Refs. [24, 25, 26], we take these to have trivial input and output systems ∣IA∣=∣OA∣=1|\mathcal{I}_A|=|\mathcal{O}_A|=1 and ∣IB∣=∣OB∣=1|\mathcal{I}_B|=|\mathcal{O}_B|=1, so that they reduce to maps fA:X→Af_A:\mathcal{X}\to\mathcal{A} and fB:Y→Bf_B:\mathcal{Y}\to\mathcal{B}.

Logical consistency translates into a unique fixed point characterization of process functions [23, 24], which, in the bipartite case, imposes the following constraints:

Theorem 1 (Characterization of bipartite process functions [24]). A bipartite process function is a pair of local functions w:=(wA,wB):A×B→X×Y\mathbf{w}:=(w_A,w_B):\mathcal{A}\times\mathcal{B}\to\mathcal{X}\times\mathcal{Y} completely characterized by

  1. No-self-signaling. The local function producing the input of a party is independent of this party’s output:

w(a,b)=(wA(b),wB(a)).(13)w(a,b)=(w_A(b),w_B(a)). \tag*{(13)}
  1. No-two-way-signaling. At least one of the local function (wA,wB)(w_A,w_B) is a constant.

We can now establish the correspondence between bipartite process functions and the FR product of unipartite Bell scenarios. The proof below does not assume any fixed cardinalities for the sets A\mathcal{A}, B\mathcal{B}, X\mathcal{X}, and Y\mathcal{Y}, and therefore applies beyond the Boolean case B2,2,2B_{2,2,2}. More generally, one could allow the number of outcomes to depend on the measurement setting. However, to establish the correspondence with process functions, it is sufficient to consider the case in which every setting has the same number of outcomes.

Proposition 1. Bipartite process functions are in one-to-one correspondence with the hyperedges of the Bell scenario B2,m,d=B1,mA,dAFR⊗B1,mB,dBB_{2,m,d}=B_{1,m_A,d_A}\mathop{\mathrm{FR}\otimes}B_{1,m_B,d_B}, where m=(mA,mB)\mathbf{m}=(m_A,m_B) and d=(dA,dB)\mathbf{d}=(d_A,d_B).

Proof. Given a hyperedge of B1,mA,dAFR⊗B1,mB,dBB_{1,m_A,d_A}\mathop{\mathrm{FR}\otimes}B_{1,m_B,d_B}, its vertices vv correspond to events (a,b∣x,y)(a,b|x,y). Since every output pair (a,b)(a,b) defining these events appears exactly once in the hyperedge, the events define a function

w=(wA,wB):A×B→X×Y,(14)\mathbf{w}=(w_A,w_B):\mathcal{A}\times\mathcal{B}\to\mathcal{X}\times\mathcal{Y}, \tag*{(14)}

where w(a,b)=(x,y)w(a,b)=(x,y) is the unique setting pair associated with the output pair (a,b)(a,b).

By the definition of the FR product (Def. 4), every hyperedge is either in EB→AE_{B\to A} or in EA→BE_{A\to B}. Consider first a hyperedge in EB→AE_{B\to A}. It is of the form

⋃v∈eA{v}×f(v),(15)\bigcup_{v\in e_A}\{v\}\times f(v), \tag*{(15)}

where eAe_A is an edge of B1,mA,dAB_{1,m_A,d_A} and f:eA→EBf:e_A\to E_B assigns a local edge of Bob’s scenario to each output of Alice. Hence, the corresponding events are (a,b∣x0,wB(a))(a,b|x_0,w_B(a)), for some fixed setting x0x_0, implying

wA(a,b)=x0,wB(a,b)=wB(a).(16)w_A(a,b)=x_0,\qquad w_B(a,b)=w_B(a). \tag*{(16)}

Similarly, a hyperedge in EA→BE_{A\to B} has the form

⋃v∈eBf(v)×{v},(17)\bigcup_{v\in e_B}f(v)\times\{v\}, \tag*{(17)}

and gives rise to events (a,b∣wA(b),y0)(a,b|w_A(b),y_0), such that

wA(a,b)=wA(b),wB(a,b)=y0.(18)w_A(a,b)=w_A(b),\qquad w_B(a,b)=y_0. \tag*{(18)}

Therefore, every hyperedge defines a pair of local functions satisfying no-self-signaling and no-two-way-signaling. By Theorem 1, the resulting map w\mathbf{w} is a bipartite process function.

Conversely, let w\mathbf{w} be a bipartite process function. By Theorem 1, it satisfies w(a,b)=(wA(b),wB(a))\mathbf{w}(a,b)=(w_A(b),w_B(a)), with at least one of the two component functions being constant. The associated set of events (a,b∣wA(b),wB(a))(a,b|w_A(b),w_B(a)) therefore has one of the forms

(a,b∣x0,y0),(a,b∣x0,wB(a)),(a,b∣wA(b),y0),(19)(a,b|x_0,y_0),\qquad(a,b|x_0,w_B(a)),\qquad(a,b|w_A(b),y_0), \tag*{(19)}

where x0x_0 and y0y_0 are constants. These are precisely the hyperedges obtained from EB→AE_{B\to A} and EA→BE_{A\to B} according to Definition 4. Hence, every bipartite process function defines a hyperedge of B1,mA,dAFR⊗B1,mB,dBB_{1,m_A,d_A}\mathop{\mathrm{FR}\otimes}B_{1,m_B,d_B}.

Bipartite process functions and hyperedges of the FR product are, therefore, equivalent. □\square

This proposition establishes a causal interpretation of the FR product, which will be generalized to arbitrary numbers of parties in the following sections. In particular, each hyperedge corresponds to a specific causal order of the parties, as characterized by the associated bipartite process function. Communication hyperedges A→BA\to B (respectively, B→AB\to A) correspond to process functions with causal order A≺BA\prec B (respectively, B≺AB\prec A), where AA lies in the causal past (respectively, BB lies in the causal past) of the other party. The no-communication hyperedges correspond to the case in which both causal orders A≺BA \prec B and B≺AB \prec A are compatible, since neither party’s input depends on the other’s output.

Moreover, the necessity of one-way communication for enforcing no-signaling (cf. Section IV A) admits a causal reformulation: to impose the bipartite no-signaling principle, one must consider all bipartite process functions, and hence all deterministic classical communications compatible with the no-self-signaling and no-two-way-signaling constraints.

The disjunctive Foulis-Randall product generates all multipartite process functions

In the bipartite case, the FR product is uniquely defined. This, however, is not true for the multipartite case. The reason is that the bipartite FR product is not associative, and hence does not extend uniquely to more than two parties. Already for three parties it holds that

HA⊗FR(HB⊗FRHC)≠(HA⊗FRHB)⊗FRHC,(20)H_A \otimes_{\mathrm{FR}} (H_B \otimes_{\mathrm{FR}} H_C) \ne(H_A \otimes_{\mathrm{FR}} H_B) \otimes_{\mathrm{FR}} H_C, \tag*{(20)}

since the two orders of composition allow different communication patterns (Proposition 3.3.1 in [9]).

In fact, a whole hierarchy of multipartite products arises [9, 15]. This leads to the minimal, common, maximal, and the disjunctive FR product. On the level of classical, quantum and boxworld models these are all equivalent, while this is not true for all probabilistic model [15]. We discuss the hierarchy in more detail in Section VI B.

To generalize Proposition 1, we can employ the nested structure of the hierarchy. All four products share the same vertex set, the Cartesian product V(H)=∏kV(Hk)V(H) = \prod_k V(H_k); they differ only in their hyperedge sets, which are nested (Proposition 1 in [15]):

E(⊗kmin⁡Hk)⊆E(⊗kcomHk)⊆E(⊗kmax⁡Hk)⊆E(⊗kdisjHk).(21)E\left(\mathop{\otimes}\limits^{\min}_k H_k\right) \subseteq E\left(\mathop{\otimes}\limits^{\mathrm{com}}_k H_k\right) \subseteq E\left(\mathop{\otimes}\limits^{\max}_k H_k\right) \subseteq E\left(\mathop{\otimes}\limits^{\mathrm{disj}}_k H_k\right). \tag*{(21)}

If we show that the hyperedges of the largest of these products, the disjunctive FR product, correspond exactly to the set of all process functions, then the hyperedges of every other product in the chain are automatically valid process functions, since they are special cases of those generated by the disjunctive FR product. Hence, we focus on the latter.

After introducing the multipartite process function framework in Section V A, and defining the disjunctive FR product in Section V B, we establish in Section V C that its hyperedges precisely characterize the set of multipartite process functions.

Multipartite process functions

Consider nn parties labeled by k∈⟦1,n⟧:={1,…,n}k \in\llbracket1,n\rrbracket:= \{1,\ldots,n\}. We define for each party kk the input variables xk∈Xkx_k \in\mathcal{X}_k,

the output variable ak∈Aka_k \in\mathcal{A}_k, as well as external inputs ik∈Iki_k \in\mathcal{I}_k and outputs ok∈Oko_k \in\mathcal{O}_k (cf. Figure 5). The local interventions of each party are modeled by local stochastic channels P(ak,ok∣xk,ik)P(a_k,o_k\mid x_k,i_k). We emphasize again that we restrict to single-round communications, i.e., each party receives, sends, and acts on a system only once. In the following, we will use the shorthand notation X:=∏kXk\mathcal{X} := \prod_k \mathcal{X}_k, A:=∏kAk\mathcal{A} := \prod_k \mathcal{A}_k, x:=(x1,…,xn)x := (x_1,\ldots,x_n), xk:=(x1,…,xk−1,xk+1,…,xn)\mathbf{x}_k := (x_1,\ldots,x_{k-1},x_{k+1},\ldots,x_n), a:=(a1,…,an)a := (a_1,\ldots,a_n) etc. The observed correlations, as specified for the bipartite case in Eq. (12), are now given by

Figure 5. Sketch of the multipartite process-function communication scenario, generalizing Figure 4 to nn parties: each party kk performs an arbitrary local intervention fkf_k, receiving the incoming variable xkx_k from the process function ww and returning the output variable aka_k. As before, the external inputs iki_k and outputs oko_k are taken to be trivial and can thus be neglected [24, 25, 26].

P(o∣i)=∑a,x(∏k=1nP(ak,ok∣xk,ik))P(x∣a),(22)P(o\mid i) = \sum_{a,x} \left(\prod_{k=1}^{n} P(a_k,o_k\mid x_k,i_k)\right) P(x\mid a), \tag*{(22)}

where P(x∣a)P(x\mid a) denotes the classical process that is carried out by the environment. Multipartite process functions are defined as follows:

Definition 6 (Multipartite process function). A process function is a set of local functions w:=(w1,…,wn):A→Xw := (w_1,\ldots,w_n) : \mathcal{A} \to\mathcal{X}, corresponding to a deterministic classical process P(x∣a)=δx,w(a)P(x\mid a) = \delta_{x,w(a)} that is logically consistent: for every choice of local operations, Eq. (22) yields a valid conditional probability distribution P(o∣i)P(o\mid i) (i.e., P(o∣i)≥0P(o\mid i) \ge0 and ∑oP(o∣i)=1\sum_o P(o\mid i) = 1 for all ii).

The process function scenario depicted in Figure 5 involves deterministic local interventions f:=(fk)k\mathbf{f} := (f_k)_k (where fk:Xk×Ik→Ak×Okf_k : \mathcal{X}_k \times\mathcal{I}_k \to\mathcal{A}_k \times\mathcal{O}_k) such that Pk(ak,ok∣xk,ik)=δ(ak,ok),fk(xk,ik)P_k(a_k,o_k\mid x_k,i_k) = \delta_{(a_k,o_k),f_k(x_k,i_k)}. As in the bipartite case, we assume that ∣Tk∣=∣Ok∣=1|\mathcal{T}_k| = |\mathcal{O}_k| = 1, yielding local interventions fk:Xk→Akf_k : \mathcal{X}_k \to\mathcal{A}_k.

There are several criteria to validate logical consistency in the multipartite case [23, 24, 26, 32]. Here, we employ a combinatorial, event-based characterization:

Theorem 2 (Pairwise-exclusive completeness [32]). A list of events E⊆A×XE \subseteq\mathcal{A} \times\mathcal{X} is the event list

{(a∣x)∣a∈A,x=w(a)}(23)\{(a|x) \mid a \in\mathcal{A}, x = w(a)\} \tag*{(23)}

of an nn-partite process function if and only if

(i) EE is complete, i.e.,

∣E∣=∣A∣=∏k∣Ak∣,(24)|E| = |\mathcal{A}| = \prod_k |\mathcal{A}_k|, \tag*{(24)}

(ii) Any two distinct events (a∣x)(a|x) and (a′∣x′)(a'|x') in EE are exclusive, i.e., there exists a party kk such that

ak≠ak′andxk=xk′.(25)a_k \ne a'_k \qquad\text{and}\qquad x_k = x'_k. \tag*{(25)}

Remarkably, for n≥3n \ge3, there exist process functions that generate correlations incompatible with any well-defined causal order. Causal process functions are indeed formalized by the following constraints [26]:

Theorem 3 (Characterization of causal process functions [26]). An nn-partite process function w:A→Xw : \mathcal{A} \to\mathcal{X} is causal if and only if the following two conditions hold:

(i) There exists at least one party kk such that no other party can signal to it, i.e. the corresponding input is constant, xk=wk(a\k):=xk0x_k = w_k(\mathbf{a}_{\backslash k}) := x_k^0.

(ii) For every such party kk and for every ak∈Aka_k \in\mathcal{A}_k, the output-reduced process functions wak:A\k→X\kw^{a_k} : \mathcal{A}_{\backslash k} \to\mathcal{X}_{\backslash k} defined as

wak(a\k):=(w1(a\1,ak),…,wk−1(a\{k−1},ak),wk+1(a\{k+1},ak),…,wn(a\n,ak))(26)w^{a_k}(\mathbf{a}_{\backslash k}) := \bigl(w_1(\mathbf{a}_{\backslash1},a_k),\ldots,w_{k-1}(\mathbf{a}_{\backslash\{k-1\}},a_k),w_{k+1}(\mathbf{a}_{\backslash\{k+1\}},a_k),\ldots,w_n(\mathbf{a}_{\backslash n},a_k)\bigr) \tag*{(26)}

are themselves valid (n−1)(n-1)-partite causal process functions.

A “non-causal” process function either has no global past or can be reduced to a function without a global past. The canonical example of a non-causal process function is the so-called Lugano (AF/BW) process [22] [29], in which each party’s input depends non-trivially on the outputs of the other two, so that all three parties lie in each other’s past and future:

x1:=a3(a2⊕1),x2:=a1(a3⊕1),x3:=a2(a1⊕1),(27)x_1 := a_3(a_2 \oplus1), \quad x_2 := a_1(a_3 \oplus1), \quad x_3 := a_2(a_1 \oplus1), \tag*{(27)}

where x1,x2,x3x_1,x_2,x_3 and a1,a2,a3a_1,a_2,a_3 label the input and output bits for the first, second, and third party, respectively.

The disjunctive Foulis-Randall product

We now recall the disjunctive FR product introduced in Ref. [15]. As discussed at the beginning of this section, it contains all the hyperedges of the FR products suggested in the original AFLS framework [9], and strictly more. Its definition proceeds in three steps. First, one needs to introduce a notion of orthogonality for composite events. This turns the set of such events into an orthogonality graph. The cliques – its sets of mutually connected vertices – are precisely the collections of pairwise exclusive events, and thus candidates for valid hyperedges of the composite contextuality scenario. Finally, a clique is considered valid if and only if it is compatible with the normalization and no-signaling conditions.

a. Vertices and orthogonality. As discussed after Definition 1, the hyperedges of any hypergraph define a notion of orthogonality, which can be formalized as follows:

Definition 7 (Orthogonality of vertices). Two distinct vertices v,w∈V(H)v,w \in V(H) are orthogonal in HH if and only if they are distinct and belong to a common hyperedge:

v⊥w  ⟺  v≠w∧∃e∈E(H):{v,w}⊆e.(28)v \perp w \iff v \ne w \land\exists e \in E(H) : \{v,w\} \subseteq e. \tag*{(28)}

Since orthogonality is a binary relation between vertices, it can be represented by a simple 1 undirected graph (see Figure 6).

Figure 6. Scheme of single- and bipartite orthogonality graph (adapted from [9]). a shows the vertex set of B1,2,2B_{1,2,2} (left) and its associated orthogonality graph (right). b shows the vertex set of B2,2,2B_{2,2,2} (left) and the associated orthogonality graph (right) which coincides with the faithful orthogonality graph (cf. Def. 10).

Definition 8 (Orthogonality graph). The orthogonality graph OrthoGraph(H)(H) associated with a hypergraph HH is the simple undirected graph with vertex set V(H)V(H) and adjacency relation

v∼w  ⟺  v⊥w.v \sim w \iff v \perp w.

Multipartite composite scenarios admits an additional and natural notion of orthogonality, that we term exclusivity. Noting that the vertex set of a composite contextuality scenario is given by the Cartesian product V(⨂k=1nHk)=∏kV(Hk)V(\bigotimes_{k=1}^{n} H_k) = \prod_k V(H_k), one may identify two composite vertices to be exclusive whenever their local components are orthogonal for at least one party:

Definition 9 (Exclusivity). Two vertices v=(v1,…,vn)v=(v_1,\ldots,v_n) and w=(w1,…,wn)w=(w_1,\ldots,w_n) of a composite hypergraph H=H1⊗⋯⊗HnH=H_1\otimes\cdots\otimes H_n are exclusive if and only if vv and ww contain local vertices vk,wkv_k,w_k that are orthogonal:

v⊥Ew  ⟺  ∃k:vk⊥wk.(29)v \perp_E w \iff\exists k : v_k \perp w_k. \tag*{(29)}

This relation is referred to as local orthogonality in Refs. [9, 15], following the multipartite principle introduced in Ref. [41]. We use the term exclusivity to emphasize that we refer only to the underlying combinatorial relation.

b. The faithful orthogonality graph. In general, the orthogonality relations between vertices that are induced by Def. 7 and Def. 9 are not identical. Exclusivity depends only on the local scenarios and not on the particular multipartite composition rule, in contrast to Def. 7, referred to as “conventional” orthogonality in [15]. Exclusivity is said to be captured by a composition, if the conventional orthogonality relations contain the exclusivity relations. If both notions lead to the same relations, then a composition is said to faithfully capture exclusivity. In particular, a composition satisfies this if and only if it is constructed using a specific graph:

Definition 10 (Faithful orthogonality graph). We define

FaithfulOrthoGraph⁡(H1,…,Hn):=OrthoGraph⁡(H1)∗⋯∗OrthoGraph⁡(Hn)(30)\operatorname{FaithfulOrthoGraph}(H_1,\ldots,H_n) := \operatorname{OrthoGraph}(H_1) * \cdots* \operatorname{OrthoGraph}(H_n) \tag*{(30)}

with the disjunctive or co-normal graphical product:

(v1,…,vn)∼(w1,…,wn) is edge in g1∗⋯∗gn⇔∃k:vk∼wk is edge in gk.(31)(v_1,\ldots,v_n) \sim(w_1,\ldots,w_n)\text{ is edge in }g_1 * \cdots* g_n \Leftrightarrow\exists k: v_k \sim w_k\text{ is edge in }g_k. \tag*{(31)}

By construction, two vertices are adjacent in FaithfulOrthoGraph⁡(H1,…,Hn)\operatorname{FaithfulOrthoGraph}(H_1,\ldots,H_n) if and only if the corresponding multipartite events are exclusive. Moreover, the disjunctive graph product is associative,

(g1∗g2)∗g3=g1∗(g2∗g3),(32)(g_1 * g_2) * g_3 = g_1 * (g_2 * g_3), \tag*{(32)}

i.e., unlike the bipartite FR product (see Eq. (20)), the faithful orthogonality graph is unambiguously defined for any number of parties.

The outcomes of any multipartite measurement must be pairwise exclusive and hence form a clique of the faithful orthogonality graph. A clique, however, specifies only a set of mutually exclusive events: pairwise exclusivity alone does not ensure that this set constitutes a complete measurement. Cliques therefore provide candidate hyperedges, and an additional condition is needed to determine which of them define valid multipartite measurements.

c. Selecting valid hyperedges. The candidate hyperedges are filtered by requiring the corresponding normalization condition to be implied by the operational constraints of the multipartite Bell scenario, namely, normalization and no-signaling. More precisely, a clique ee defines a hyperedge when the condition ∑v∈ep(v)=1\sum_{v\in e}p(v)=1 is a linear combination of (i) the normalization conditions, one for each choice of a local measurement made by every party independently (the non-communicating, Cartesian-product measurements), and (ii) the no-signaling equalities,

∑vk∈ekp(…,vk,…)=∑vk∈ek′p(…,vk,…),(33)\sum_{v_k\in e_k}p(\ldots,v_k,\ldots)=\sum_{v_k\in e'_k}p(\ldots,v_k,\ldots), \tag*{(33)}

for every party kk, every pair of local hyperedges ek,ek′∈E(Hk)e_k,e'_k\in E(H_k), and fixed events of the remaining parties. Equivalently, this condition can be formulated as a rank comparison of the associated {0,±1}\{0,\pm1\}-valued matrices [15]; we summarize the corresponding matrix formulation in Appendix B.

The above criterion specifies which pairwise-exclusive (in the sense of Def. 7) sets of events define valid measurements. In particular, it provides the hyperedge structure of the multipartite scenario introduced in Ref. [15], namely the disjunctive FR product.

Definition 11 (Disjunctive FR product [15]). The disjunctive FR product \mathop{\bigcirc\limits^{\mathrm{disj}}}_{k=1}^{n}H_k has vertex set ∏k=1nV(Hk)\prod_{k=1}^{n}V(H_k) and, as hyperedges, exactly those cliques

of FaithfulOrthoGraph(H1,…,Hn)FaithfulOrthoGraph(H_1,\ldots,H_n) whose normalization condition follows from the normalization and no-signaling constraints.

Observation 1. Importantly, for the case of unipartite Bell scenarios B1,mk,dkB_{1,m_k,d_k}, the notion of exclusivity of Definition 7 coincides with Eq. (25).

Proof. For Hk=B1,mk,dkH_k = B_{1,m_k,d_k}, composite vertices of H=H1⊗⋯⊗HnH = H_1 \otimes\cdots\otimes H_n are represented by events (a∣x):=(a1,…,an∣x1,…,xk)(a|x) := (a_1,\ldots,a_n|x_1,\ldots,x_k). Two events (a∣x),(a′∣x′)(a|x),(a'|x') are exclusive, if and only if there are local events (ak∣xk),(ak′∣xk′)(a_k|x_k),(a'_k|x'_k) that are orthogonal according to Def. 7. This enforces that they are in the same local hyperedge, i.e., xk=xk′x_k = x'_k, as well as ak≠ak′a_k \ne a'_k, since otherwise the two local events are identical and thus not orthogonal according to Def. 7. □

Notably, the disjunctive FR product of unipartite Bell scenarios contains hyperedges for which no party is required to measure first. Ref. [15] provided indeed an explicit example of such a hyperedge, for the case of multipartite Bell scenarios disj⊗k=1nB1,mk,dk\mathop{\mathrm{disj}}\limits_{\otimes k=1}^{n} B_{1,m_k,d_k}:

E={(000∣001),(001∣101),(010∣100),(011∣100),(100∣001),(101∣111),(110∣101),(111∣111)}.(34)E = \{(000|001),(001|101),(010|100),(011|100),\\ (100|001),(101|111),(110|101),(111|111)\}. \tag*{(34)}

We identify these events as the ones constituting the following Lugano-like process2

x1=(a2⊕1)(a3⊕1)⊕1,x2=a3a1,x3=(a1⊕1)a2⊕1(35)\begin{aligned} x_1 &= (a_2 \oplus1)(a_3 \oplus1) \oplus1,\\ x_2 &= a_3a_1,\\ x_3 &= (a_1 \oplus1)a_2 \oplus1 \tag*{(35)} \end{aligned}

In the next section, we show that the hyperedges of the disjunctive FR product are in one-to-one correspondence with multipartite process functions.

Equivalence of process functions and disjunctive Foulis-Randall hyperedges

We now come to our main result: the hyperedges of the disjunctive FR product of unipartite Bell scenarios are exactly the process functions. The proof rests on two ingredients.

On the process-function side, Theorem 2 characterizes them combinatorially: an event list E⊂A×XE \subset\mathcal{A} \times\mathcal{X} is the event list {(a∣w(a))}a∈A\{(a|w(a))\}_{a\in\mathcal{A}} of an nn-partite process function if and only if it is complete and pairwise exclusive.

On the hypergraph side, we reformulate the validity criterion of the disjunctive FR product on unipartite Bell scenarios – which demands that the normalization of a hyperedge is a linear combination of the normalization and no-signaling conditions – as the requirement that the probabilities of its events sum to one for every no-signaling model (Lemma 1, Eq. (39)).

We will then show (Theorem 4) that this validity criterion is precisely pairwise exclusive completeness.

Consider the vertex set of composite events

V=⨂k=1nV(B1,mk,dk)={(a∣x):a∈A,x∈X}.(36)V = \bigotimes_{k=1}^{n} V(B_{1,m_k,d_k}) = \{(a|x):a\in\mathcal{A},x\in\mathcal{X}\}. \tag*{(36)}

and we are interested in which composition of ⨂k=1nB1,mk,dk\bigotimes_{k=1}^{n} B_{1,m_k,d_k}, i.e. which partition of VV as hyperedges, corresponds to the whole set of process functions. The normalization and no-signaling conditions can therefore be thought, in general, as conditions on assignments p∈RVp\in\mathbb{R}^{V}, independent of any choice of hyperedges. They can be written as

Nx(p):=∑a∈Ap(a∣x)−1,(37)N_x(p):=\sum_{a\in\mathcal{A}}p(a|x)-1, \tag*{(37)}
NSi(p):=∑akp(aka∣k∣xkx∣k)−∑akp(aka∣k∣xk′x∣k),(38)NS_i(p):=\sum_{a_k}p(a_k a_{\lvert k} | x_k x_{\lvert k})-\sum_{a_k}p(a_k a_{\lvert k}|x'_k x_{\lvert k}), \tag*{(38)}

where Eq. (37) is indexed by x∈Xx\in\mathcal{X} and Eq. (38) by i=(k,xk,xk′,x∣k,a∣k)i=(k,x_k,x'_k,x_{\lvert k},a_{\lvert k}), expressing that the marginal of all parties but kk does not depend on party kk's setting. A no-signaling model is an assignment p:V→[0,1]p:V\to[0,1] with Nx(p)=0N_x(p)=0 and NSi(p)=0NS_i(p)=0 for all xx and ii, i.e. a probabilistic model (Def. 3) on the vertex set Eq. (36).

Lemma 1. A set of events e⊂Ve\subset V is a valid hyperedge of disj⊗k=1nB1,mk,dk\mathop{\mathrm{disj}}\limits_{\otimes k=1}^{n}B_{1,m_k,d_k} if and only if

∑v∈ep(v)=1for every no-signaling model p.(39)\sum_{v\in e}p(v)=1\qquad\text{for every no-signaling model }p. \tag*{(39)}

Proof. (⇒)(\Rightarrow) Let ee be a disjunctive FR hyperedge of disj⊗k=1nB1,mk,dk\mathop{\mathrm{disj}}\limits_{\otimes k=1}^{n}B_{1,m_k,d_k}. By Def. 11, its normalization is a linear combination of the conditions (37)–(38): there exist coefficients λx,μi∈R\lambda_x,\mu_i\in\mathbb{R} such that

∑v∈ep(v)−1=∑xλxNx(p)+∑iμiNSi(p),∀p∈RV.(40)\sum_{v\in e}p(v)-1=\sum_x\lambda_xN_x(p)+\sum_i\mu_iNS_i(p),\qquad\forall p\in\mathbb{R}^{V}. \tag*{(40)}

If pp is a no-signaling model, by definition, every term on the right-hand side vanishes, which gives Eq. (39).

(⇐)(\Leftarrow) Let e⊂Ve\subset V satisfy Eq. (39). We verify the two requirements of Def. 11.

(i)(i) ee is a clique of the orthogonality graph. Recall that a local deterministic model is specified by the choice of a single response function fk:Xk→Akf_k:\mathcal{X}_k\to\mathcal{A}_k per party, where each party kk outputs fk(xk)f_k(x_k) on setting xkx_k with certainty, i.e., for a fixed string of measurement settings xx, only one joint outcome aa appears with certainty:

pf(a∣x)=∏k=1n[ak=fk(xk)]∈{0,1}.(41)p_f(a|x)=\prod_{k=1}^{n}[a_k=f_k(x_k)]\in\{0,1\}. \tag*{(41)}

Such models are local, hence no-signaling, so Eq. (39) applies to them.

Assume, ad absurdum, that two distinct events (a∣x),(a′∣x′)∈e(a|x),(a'|x') \in e are not exclusive. By Observation 1, exclusivity would require a party kk with xk=xk′x_k=x'_k and ak≠ak′a_k\ne a'_k; its failure therefore means that for every party kk such that xk=xk′x_k=x'_k, we have ak=ak′a_k=a'_k, and thus, in this case, fk(xk)=fk(xk′)=ak=ak′f_k(x_k)=f_k(x'_k)=a_k=a'_k. Choosing the responses to the remaining settings arbitrarily, we obtain a local deterministic model Eq. (41) that assigns probability 1 to both events, pf(a∣x)=pf(a′∣x′)=1p_f(a|x)=p_f(a'|x')=1, so that ∑v∈epf(v)≥2\sum_{v\in e}p_f(v)\ge2, contradicting Eq. (39). Hence the events of ee are pairwise exclusive.

(ii) The normalization of ee follows from (37)–(38). These conditions cut out the affine subspace A={p∈RX:Nx(p)=NSi(p)=0 ∀x,i}\mathsf{A}=\{p\in\mathbb{R}^{\mathcal{X}}:N_x(p)=\mathrm{NS}_i(p)=0\ \forall x,i\}, whose elements need not be non-negative; the no-signaling models are precisely those that are, NS={p∈A:p≥0}\mathcal{NS}=\{p\in\mathsf{A}:p\ge0\}. By Eq. (39), the affine map p↦∑v∈ep(v)−1p\mapsto\sum_{v\in e}p(v)-1 becomes the zero-map on NS\mathcal{NS}, hence on all of A\mathsf{A}, since NS\mathcal{NS} affinely spans A\mathsf{A} – as the local models already do [42]. As p↦∑v∈ep(v)−1p\mapsto\sum_{v\in e}p(v)-1 vanishes on NS\mathcal{NS} – which is by definition the solution set of Nx(p)=NSi(p)=0 ∀x,i, ∀p≥0N_x(p)=\mathrm{NS}_i(p)=0\ \forall x,i,\ \forall p\ge0 – it is a linear combination of these equations. Hence Eq. (40) holds. Thus ee is a valid disjunctive FR hyperedge.

As we now show, Eq. (39) is precisely the structure of pairwise exclusive completeness3.

Theorem 4. There is a one-to-one correspondence between hyperedges of the disjunctive FR product of unipartite Bell scenarios (Def. 11) and process functions (Def. 6).

Proof. Throughout, pfp_f denotes the local deterministic model (41) associated with a tuple of response functions f=(f1,…,fn)∈F:=∏k=1nAkXkf=(f_1,\ldots,f_n)\in\mathcal{F}:=\prod_{k=1}^n\mathcal{A}_k^{\mathcal{X}_k}, where AkXk\mathcal{A}_k^{\mathcal{X}_k} denotes the set of functions fk:Xk→Akf_k:\mathcal{X}_k\to\mathcal{A}_k, and we abbreviate ∣A∣=∏k=1n∣Ak∣|\mathcal{A}|=\prod_{k=1}^n|\mathcal{A}_k|.

(⇒)(\Rightarrow) Let ee be a hyperedge of disj⁡⊗k=1nB1,mk,dk\operatorname{disj}\otimes_{k=1}^nB_{1,m_k,d_k}. By definition of the disjunctive FR product, ee is a clique of the orthogonality graph, i.e. a set of pairwise exclusive events (Observation 1). Let us show that it is also complete. By Lemma 1 the normalization Eq. (39) holds for every no-signaling model, in particular for the constant local deterministic models pfap_{f_a}, a∈Aa\in\mathcal{A}, defined by fka(xk)=akf_k^a(x_k)=a_k for all xk∈Xkx_k\in\mathcal{X}_k: each party kk outputs aka_k irrespective of its setting. Such a model assigns probability 1 to the events of ee with outcome tuple aa and 0 to all others, so that for all pfap_{f_a},

1=∑v=(a′∣x′)∈epfa(a′∣x′)=∑(a′∣x′)∈eδa′,a.1=\sum_{v=(a'|x')\in e}p_{f_a}(a'|x')=\sum_{(a'|x')\in e}\delta_{a',a}.

Every outcome tuple a∈Aa\in\mathcal{A} thus occurs in exactly one event of ee. Thus ∣e∣=∣A∣|e|=|\mathcal{A}| and e={(a∣w(a))}a∈Ae=\{(a|w(a))\}_{a\in\mathcal{A}} for a unique map w:A→Xw:\mathcal{A}\to\mathcal{X}. Being complete and pairwise exclusive, ee is therefore the event list of a process function ww by Theorem 2.

(⇐)(\Leftarrow) Let ww be a process function, defining the complete pairwise exclusive event list Sw={(a∣w(a))}a∈A\mathcal{S}_w=\{(a|w(a))\}_{a\in\mathcal{A}}. By Ref. [26] (Section VII B) the resulting pairwise exclusive events obey a trivial Bell inequality [43, 44], i.e. an inequality not violated by any no-signaling correlation – equivalently, a trivial local orthogonality inequality [41]:

∑v∈Swp(v)≤1for every no-signaling model p.(42)\sum_{v\in\mathcal{S}_w}p(v)\le1\qquad\text{for every no-signaling model }p. \tag*{(42)}

It remains to show that Eq. (42) is in fact saturated by every no-signaling model, since Lemma 1 then identifies Sw\mathcal{S}_w as a hyperedge of disj⁡⊗k=1nB1,mk,dk\operatorname{disj}\otimes_{k=1}^nB_{1,m_k,d_k}.

Averaging the local deterministic models defined in Eq. (41) over all f∈Ff\in\mathcal{F} yields the uniform model,

1∣F∣∑f∈Fpf=p∗,p∗(a∣w(a))=1∣A∣.\frac{1}{|\mathcal{F}|}\sum_{f\in\mathcal{F}}p_f=p_*,\qquad p_*(a|w(a))=\frac{1}{|\mathcal{A}|}.

Since Sw\mathcal{S}_w is complete, ∣Sw∣=∣A∣|\mathcal{S}_w|=|\mathcal{A}| and therefore ∑v∈Swp∗(v)=1\sum_{v\in\mathcal{S}_w}p_*(v)=1. As the sum over Sw\mathcal{S}_w is linear in the model,

1∣F∣∑f∈F∑v∈Swpf(v)=∑v∈Swp∗(v)=1,\frac{1}{|\mathcal{F}|}\sum_{f\in\mathcal{F}}\sum_{v\in\mathcal{S}_w}p_f(v)=\sum_{v\in\mathcal{S}_w}p_*(v)=1,

while ∑v∈Swpf(v)≤1\sum_{v\in\mathcal{S}_w}p_f(v)\le1 for every ff by Eq. (42). An average of numbers bounded by 1 can equal 1 only if each of them does, so ∑v∈Swpf(v)=1\sum_{v\in\mathcal{S}_w}p_f(v)=1 for every local deterministic model. As these affinely span the subspace A\mathsf{A} defined by the normalization and no-signaling conditions [42], the affine map p↦∑v∈Swp(v)−1p\mapsto\sum_{v\in\mathcal{S}_w}p(v)-1 vanishes on all of A\mathsf{A}, hence in particular on every no-signaling model. Thus Sw\mathcal{S}_w satisfies Eq. (39), and Lemma 1 shows that it is a hyperedge of disj⁡⊗k=1nB1,mk,dk\operatorname{disj}\otimes_{k=1}^nB_{1,m_k,d_k}.

In the following sections, we explore several implications of Theorem 4.

Applications

No-signaling based characterisation of process functions

Theorem 4 unveils a connection between two seemingly separate concepts: process functions and measurement contexts. Specifically, a deterministic classical process given by

P(x∣a)={1x=w(a)0otherwise(43)P(x|a) = \begin{cases} 1 & x = w(a) \\ 0 & \text{otherwise} \end{cases} \tag*{(43)}

determines a measurement context ee consisting of events {(a∣x)∣P(x∣a)=1}\{(a|x) \mid P(x|a) = 1\}. In particular, the properties of ee and the properties of process functions are equivalent:

Corollary 1. Logical consistency of process functions (Def. 6) is equivalent to the validity criterion of a hyperedge of the disjunctive FR product of unipartite Bell scenarios (Def. 11).

Theorem 4 then implies a new necessary and sufficient criterion for the logical consistency of process functions. Whereas the characterization of Theorem 2 ([32]) rests on completeness and pairwise exclusivity, the following is phrased entirely in terms of no-signaling and normalization conditions.

Corollary 2. A function w:A→Xw : \mathcal{A} \to\mathcal{X} is a process function if and only if the normalization of its event set e:={(a∣w(a))}e := \{(a|w(a))\}, namely ∑v∈ep(v)=1\sum_{v \in e} p(v) = 1, is a linear combination of the no-signaling and normalization conditions (cf. Section V B, paragraph c); an explicit matrix-based criterion is given in Corollary 5.

Identifying Foulis-Randall products with process function types

Recall that there is no unique way to compose more than two scenarios, leading to a multitude of FR products (as discussed at the beginning of Section V). Since the largest of these products already corresponds to process functions (Theorem 4), the same holds for every product in the chain.

Observation 2. For Bell scenarios Hk=B1,mk,dkH_k = B_{1,m_k,d_k}, the hyperedges of the minimal, common, and maximal FR products are process functions.

In particular, the nesting in Eq. (21) induces a hierarchy of process functions: different FR products correspond to distinct types of process functions. Refs. [9, 15] already offered an intuition for the first three products, reading their hyperedges as parties acting within measurement protocols (the maximal product entailing all of them). The disjunctive FR product, however, contains precisely the additional valid hyperedges that no such protocol can account for. Our correspondence to process functions sharpens this picture into a genuine causal interpretation: each product in the chain of Eq. (21) corresponds to a distinct class of causal structures, ranging from process functions with fixed causal order to dynamical control of causal order, and finally to non-causality for the disjunctive FR product. We analyze the four levels in turn (Figure 7). Moreover, we compare each level with an existing hierarchy for quantum circuits [33, 45].

Minimal Foulis-Randall product

This FR product imposes an “all-for-one” communication structure, where (n−1)(n - 1) non-communicating parties all send a signal to the remaining party laying in their global future (central part of Figure 7).

The edge set is defined in analogy to the bipartite FR product (cf. Eq. (10)) [15]:

E(⨂k=1nHk)=⋃kEk=⋃k⋃v∖k{vk}×f(vk),(44)E\left(\bigotimes_{k=1}^{n} H_k\right) = \bigcup_k E_k = \bigcup_k \bigcup_{\mathbf{v}\setminus k} \{\mathbf{v}_k\} \times f(\mathbf{v}_k), \tag*{(44)}

i.e., the set of hyperedges is the union of sets of measurements EkE_k, where each party i≠ki \ne k conducts a measurement eie_i, announces the outcomes viv_i to a special party kk, which then conducts the measurement as a function of the announced outcomes, i.e., ek=f(vk)e_k = f(\mathbf{v}_k).

Considering the case of local unipartite Bell scenarios, the process functions corresponding to the hyperedges defined in Eq. (44) are of the general form

w(a)=(c1,c2,…,ck−1,wk(ak),ck+1,…,cn)(45)w(a) = (c_1, c_2, \ldots, c_{k-1}, w_k(a_k), c_{k+1}, \ldots, c_n) \tag*{(45)}

for constants ci∈Xic_i \in\mathcal{X}_i, i≠ki \ne k, i.e., all components are constant except the kkth, which can depend on all outputs except its own.

In the quantum circuit hierarchy [33], up to a relabeling of the last party, this class of process functions can be read as the deterministic and classical version of quantum circuit with operations used in parallel (QC-PAR). We prove it explicitly in Appendix C.

Common Foulis-Randall product

As remarked in [9] the definition of the minimal product arbitrarily singles out one party to conduct its measurement as a function of all the others. One may instead consider so-called correlated measurements [15]: one party initiates the protocol, chooses a measurement and communicates its outcome to all the other parties. The remaining parties then proceed in turn, each choosing a setting that may depend on the outcomes of all the parties preceding it. One may describe this protocol as imposing a “causal memory” (cf. Figure 7, second innermost layer). The edge set of the common FR product contains all possible correlated measurements.

Considering the case of local unipartite Bell scenarios, the process functions corresponding to the hyperedges of the common FR product take the general form

w(a)=(c1,w2(a1),w3(a1,a2),…,wn(an−1))(46)w(a) = (c_1, w_2(a_1), w_3(a_1,a_2), \ldots, w_n(a_{n-1})) \tag*{(46)}

where c1∈X1c_1 \in\mathcal{X}_1 is a constant, corresponding to the case where party 1 initiates the protocol. In the quantum circuit hierarchy [33], this class of process functions corresponds to the deterministic and classical version of quantum circuit with fixed causal order (QC-FO), also known as quantum combs [46, 47]. We prove it explicitly in Appendix C.

Maximal Foulis-Randall product

So far, the order of the parties in the protocol is fixed. The maximal FR product extends it to any measurement protocol compatible with a well-defined causal structure. This notably includes “dynamical” control of causal orders [45]: the causal order between parties can be controlled by the measurement outcome – and thus, as (fk(xk):=ak)(f_k(x_k) := a_k), the choice of operation – of another party. The canonical example is the classical switch, e.g. x1:=0x_1 := 0, x2:=a1a3x_2 := a_1a_3, x3:=(a1⊕1)a3x_3 := (a_1 \oplus1)a_3. Here, the measurement outcome a1a_1 for setting x1:=0x_1 := 0 of party A1A_1 controls the order of the other two parties: if a1=0a_1 = 0, A2A_2 precedes A3A_3, and if a1=1a_1 = 1, A3A_3 precedes A2A_2.

To define the edge set of the maximal FR product, [9] introduced a recursive definition of a measurement protocol:

Definition 12 (Measurement protocol [9]). A measurement protocol P\mathcal{P} for a set of parties SS consists of the following data:

  • if S={∅}S = \{\varnothing\}, the unique protocol is P=∅\mathcal{P} = \varnothing

  • otherwise, the protocol is a triple P=(k,e,f)\mathcal{P} = (k,e,f) for some party k∈Sk \in S, a hyperedge e∈E(Hk)e \in E(H_k), and a function ff assigning to each v∈ev \in e a measurement protocol f(v)f(v) on S\{k}S \backslash\{k\}

As described in [9, 15], this definition can be understood as one party kk initiating the protocol, choosing an initial measurement ee, and using a function ff to assign to every outcome vv of another measurement protocol that only includes the other parties. For instance, if the outcome of the initial measurement is v1v_1, then party ll may be the next initiator, but if the outcome is v2≠v1v_2 \ne v_1, a different party mm may be the next initiator – or the same, reproducing the fixed order-case of the common FR product. Each of these subsequent initiators can again control which party follows. One may think of the measurement protocol as a tree where the leaves are created by the different outcomes of the initiating parties (cf. Figure 8). Within one measurement, one can thus have various orders of parties (cf. third layer in Figure 8). Note that every measurement constructed in this way has again a global past, the initiating party, while the global future is in general not fixed.

Figure 8. Illustration of a measurement protocol. Considering four parties A1A_1 to A4A_4 with A1A_1 initiating the protocol, the outcomes of the measurement chosen by A1A_1 determine subprotocols, creating leaves and eventually, due to the recursive nature of Def. 12, an entire directed tree. This measurement protocol is an example of dynamical control of definite causal order.

More formally, the maximal FR product has a set of edges

E(⨂i=1nHimax⁡)=⋃PO(P).(47)E\left(\mathop{\bigotimes}\limits_{i=1}^{n} H_i^{\max}\right)=\bigcup_{\mathcal{P}} O(\mathcal{P}). \tag*{(47)}

where O(P)O(\mathcal{P}) is the set of outcomes of the measurement protocol P\mathcal{P}.

Considering the case of local Bell scenarios, if party k=1k=1 initiates the protocol, the corresponding process functions take the general form

w(a)=(c1,w2(a\2),…,wn(a\n))(48)w(\boldsymbol{a})=(c_1,w_2(\boldsymbol{a}_{\backslash2}),\ldots,w_n(\boldsymbol{a}_{\backslash n})) \tag*{(48)}

for some constant measurement setting c1c_1. No-self-signaling of the other components follows from the definition of a measurement protocol: once a party has been chosen, it cannot be selected again at a later stage. In particular, all such process functions are causal:

Proposition 2. The hyperedges of the maximal FR are equivalent to all causal process functions.

Proof. Due to Corollary 2, a hyperedge of the maximal FR product corresponds to a process function ww. It is also causal: by definition of a measurement protocol, one party initiates and thus has a constant measurement setting, i.e., one of the process function components is constant and condition (i) in Theorem 3 is satisfied. Condition (ii) is also satisfied since it mimics the structure of a measurement protocol.

Conversely, a causal process function defines events (a∣w(a))(a|w(a)) where (i) implies that there is one party in the global past, which takes the role of the first party in the measurement protocol. Condition (ii) again is exactly the definition of a measurement protocol.

In the quantum circuit hierarchy [33], this class of process functions corresponds to the deterministic and classical version of quantum circuit with classical control of causal order (QC-CC). We prove it in Appendix C. Moreover, Ref. [45] introduced the notion of quantum circuits with non-influenceable classical control of causal order (QCNICCs), defined as “the QC-CCs for which there exists an implementation such that, at any intermediate time step, the state of the control system is independent of the choice of previously applied external operations and of the potential initial state preparation in the global past.” We also show in (Appendix C, Corollary 6) that in the special case of process functions, i.e. deterministic classical QC-CC, the class of QCNICC does not exist, i.e. all deterministic classical QC-CC that do not belong to QC-FO are influenceable.

Disjunctive Foulis-Randall product

Process functions that correspond to hyperedges of the disjunctive FR product have the general form

w(a)=(w1(a\1),w2(a\2),…,wn(a\n)).(49)w(\boldsymbol{a})=(w_1(\boldsymbol{a}_{\backslash1}),w_2(\boldsymbol{a}_{\backslash2}),\ldots,w_n(\boldsymbol{a}_{\backslash n})). \tag*{(49)}

In particular, the disjunctive FR product contains hyperedges that correspond to non-causal process functions (cf. Eq. (35)). We can use this property to establish a witness of classical non-causality, which is implied by Proposition 2:

Corollary 3. If a hyperedge is contained in the disjunctive but not in the maximal FR product, it corresponds to a non-causal process function.

This class of process functions has no counterpart in the quantum circuit hierarchy [33]. The largest class of that hierarchy, quantum circuits with quantum control of causal order (QC-QCs), admits no deterministic classical analogue, since the control system is itself quantum. Conversely, process functions—the deterministic classical special case of process matrices—are not contained in the QC-QCs either: they can generate correlations violating causal inequalities, which QC-QCs cannot [33].

Generating process functions via no-signaling conditions

In the following, we show that the normalization and no-signaling conditions characterizing process functions can yield rules that permit to substitute the events of a hyperedge. These allow to generate non-causal process

functions from no-communication ones as we illustrate with the derivation of the Lugano process.

Consider the initial hyperedge that fixes every setting to x=y=z=0x=y=z=0 (Figure 9a), corresponding to the constant, no-communication, process function

Figure 9. Graphical derivation of the Lugano process. This provides a joint graphical representation of classical non-causality and non-locality without entanglement (NLWE). (a) A three-partite Bell contextuality scenario, where for each panel, the measurement setting and outcome of AA is fixed. The bars indicate directions in which the measurement settings and outcomes of party BB (green) and CC (cyan) vary. A hyperedge with constant setting for all three parties is shown in gray. (b) Applying a no-signaling substitution rule (corresponding to a condition of no-signaling from CC to AA and BB) yields a process function with communication from AA and BB to CC. (c) Applying another no-signaling substitution rule (corresponding to a condition of no-signaling from BB to AA and CC) yields a process function where AA controls the order of BB and CC, i.e., we obtain a classical switch (or rather, dynamical control of causal order). (d) Finally, applying a no-signaling condition of no-signaling from AA to BB and CC) yields a non-causal process: the Lugano process [23]. Since this hyperedge corresponds to a SHIFT-type measurement [40], which exhibits NLWE, we obtain a joint graphical representation of NLWE and classical non-causality.

w(a)=(wA(b,c),wB(a,c),wC(a,b))=(0,0,0).(50)w(\mathbf{a})=(w_A(b,c),w_B(a,c),w_C(a,b))=(0,0,0). \tag*{(50)}

The no-signaling conditions can be read directly off the hypergraph: in Figure 9a, the rows encode no-signaling from CC to A,BA,B (the local events of A,BA,B are fixed while CC’s setting varies), the columns encode no-signaling from BB to A,CA,C, and events identified across panels encode no-signaling from AA to B,CB,C (Figure 10). More formally, the no-signaling condition preventing CC from signaling to AA and BB reads

Figure 10. Three-dimensional visualization of the no-signaling substitution rule that enables the derivation of the hyperedge corresponding to the Lugano process in Figure 9d from the hyperedge in Figure 9c.

∑vC∈eCp(vA,vB,vC)=∑vC∈eC′p(vA,vB,vC),(51)\sum_{v_C\in e_C}p(v_A,v_B,v_C)=\sum_{v_C\in e'_C}p(v_A,v_B,v_C), \tag*{(51)}

for fixed vA∈V(HA)v_A\in V(H_A), vB∈V(HB)v_B\in V(H_B). Read as a substitution rule, it allows the events on the left to be replaced by those on the right.

  • a. Step 1: Applying the rule for CC not signaling to A,BA,B (Eq. (51) with vA=(0∣0)v_A=(0|0), vB=(1∣0)v_B=(1|0) and z=0,1z=0,1),

p(010∣000)+p(011∣000)=p(010∣001)+p(011∣001),(52)p(010|000)+p(011|000)=p(010|001)+p(011|001), \tag*{(52)}

yields the hyperedge of Figure 9b—obtained by “bending” the lower part of Figure 9a to the right. The process function becomes

w(a)=(0,0,a⋅b).(53)w(\mathbf{a})=(0,0,a\cdot b). \tag*{(53)}

Step 2: Applying the rule for BB not signaling to A,CA,C gives Figure 9c: AA now holds a fixed setting whose outcome controls the order of BB and CC — a dynamical control of causal order,

w(a)=(0,a(c⊕1),b(a⊕1)).(54)w(\mathbf{a})=(0,a(c\oplus1),b(a\oplus1)). \tag*{(54)}

Step 3: Finally, applying the rule for AA not signaling to B,CB,C (visualized along the third dimension in Figure 10) gives Figure 9d, in which every party’s setting varies. The result is the Lugano process,

w(a)=(c(b⊕1),a(c⊕1),b(a⊕1)).(55)w(\mathbf{a})=(c(b\oplus1),a(c\oplus1),b(a\oplus1)). \tag*{(55)}

Note that one could choose different no-signaling conditions (e.g., Eq. (51) for vA=(0∣0)v_A=(0|0), vB=(0∣0)v_B=(0|0) which in the first step yields the process function w(a)=(0,0,(a⊕1)⋅(b⊕1))w(\mathbf{a})=(0,0,(a\oplus1)\cdot(b\oplus1))). This, however, can always produce non-causal process functions that are equivalent to the Lugano process.

One can see that each step preserves logical consistency (as formalized below in Lemma 2), since every intermediate hyperedge is itself a valid process function, tracing a path through the causal hierarchy from a fixed order (Step 1, minimal FR product) to dynamical control (Step 2, maximal FR product) to non-causality (Step 3, disjunctive FR product). The only FR product not represented in this derivation is the common FR product that is in between the minimal and maximal FR product. For completeness, we provide an example for a hyperedge of the common FR product in Figure 11. This hyperedge defines the process function

Figure 11. Example of a hyperedge that is in the common Foulis-Randall product.

w(a)=(wA(b,c),wB(a,c),wC(a,b))=(0,a,a⋅b).(56)w(\mathbf{a})=(w_A(b,c),w_B(a,c),w_C(a,b))=(0,a,a\cdot b). \tag*{(56)}

Since all higher-dimensional non-causal process functions admit an AFLS representation, we expect that they can be derived via a similar construction. In particular, the no-signaling conditions are defined for any number of parties and any input and output dimension. We can read these conditions as the exchange of two local settings and refer to them as no-signaling substitution rules. Given the equivalence between process functions and hyperedges (Theorem 4), these rules can be read as substitutions of the hyperedge events.

Crucially, no-signaling substitution rules preserve the defining properties of process functions and hyperedges: exclusivity and completeness.

Lemma 2. If a no-signaling substitution rule is applicable to a hyperedge of the disjunctive FR product of Bell scenarios, then it preserves exclusivity and completeness.

Proof. This proof is inspired by proof ideas of Conjecture 2 in [15]. A no-signaling substitution rule exchanges s={(ak,aℓ∣xk,xℓ)}ak∈Aks=\{(a_k,a_\ell|x_k,x_\ell)\}_{a_k\in A_k} by s′={(ak,aℓ∣xk′,xℓ)}ak∈Aks'=\{(a_k,a_\ell|x'_k,x_\ell)\}_{a_k\in A_k}. It can be applied to a hyperedge ee if ss is contained in ee. We show that exclusivity can never be replaced by such a no-signaling substitution rule. If it could, then exclusivity of a composite event v~∈e\s\tilde{v} \in e\backslash s to v∈sv \in s would be only due to party kk (if party j≠kj \ne k would guarantee exclusivity between v~\tilde{v} and vv, it would also guarantee exclusivity for v~\tilde{v} and w∈s′w \in s', since vv and ww only differ for party kk). This would imply that x~k=xk\tilde{x}_k = x_k and a~k≠ak\tilde{a}_k \ne a_k. Now, there is also an event v′∈sv' \in s with (a~k∣xk)(\tilde{a}_k|x_k) since ss contains all ak∈Aka_k \in A_k. Then v~\tilde{v} and v′v' are exclusive due to some j≠kj \ne k. Since all events in ss share the same local events (aj∣xj)(a_j|x_j) for j≠kj \ne k, party jj also makes v~\tilde{v} exclusive with respect to vv, contradicting the assumption that only party kk is responsible for this. Hence exclusivity is preserved.

Finally, completeness is preserved: a substitution rule for party kk exchanges ∣Ak∣|A_k| events ss of ee by ∣Ak∣|A_k| events s′s' that are not in ee: if (ak,a\k∣xk′,x\k)∈s′(a_k,a_{\backslash k}|x'_k,x_{\backslash k}) \in s' were contained in ee, then it and (ak,a\k∣xk,x\k)∈s(a_k,a_{\backslash k}|x_k,x_{\backslash k}) \in s would not be exclusive, since their outputs are the same. □

Since complete pairwise-exclusive event sets are exactly process functions (Theorem 2) and hyperedges (Theorem 4), no-signaling substitution rules, when applicable, map process functions to process functions. Equivalently, they map hyperedges of the disjunctive FR product of unipartite Bell scenarios to each other.

Note that in [15] (Conjecture 2) it was mentioned that no-signaling conditions can be understood as substitution rules to generate so-called virtual hyperedges (cf. Appendix A), which inspired the construction in this section.

Discussion

In this work, we provide a correspondence between two a priori unrelated objects: the hyperedges generated by Foulis–Randall (FR) products in the Acín-Fritz-Leverrier-Sainz (AFLS) hypergraph formalism for composite Bell scenarios, and process functions — deterministic, paradox-free communication protocols that may be incompatible with any definite causal order. More precisely, the disjunctive FR product generates exactly all process functions, causal and non-causal alike, for any finite number of parties, settings, and outcomes. Logical consistency of a process function and validity of a multipartite measurement context are therefore one and the same condition.

Pairwise exclusive completeness was recently identified as a complete characterization of process functions [32]. It can also be read as a structural version [48] of Specker’s principle (“if you have several questions and you can answer any two of them, then you can also answer all of them”) [49, 50] in Bell scenarios or, equivalently, of the local orthogonality principle, introduced to recover the set of quantum correlations from the no-signaling ones [41]. Within the AFLS framework, this condition takes a particularly simple form: two events are exclusive exactly when they are distinct vertices of a common hyperedge of the Bell scenario Bn,k,mB_{n,k,m}, so that the complete pairwise exclusive sets are precisely the hyperedges in a Bell scenario. The combinatorics of Bell scenarios thus encodes every logically consistent form of deterministic classical communication, causal or non-causal alike.

This yields a causal reading of several previously introduced notions. For instance, those measurements generated by the disjunctive FR product that so far have been regarded as non-standard measurement protocols, are exactly the non-causal process functions, while each of the remaining FR products yields the process functions compatible with a specific class of definite causal orders. The resulting hierarchy can in turn be compared with that of quantum circuits with quantum control causal order (QC-QC) [33], the disjunctive FR product being the only introduced product without a quantum-circuit counterpart. The same causal reading explains why enforcing no-signaling for composite measurements requires communication. For instance, in the bipartite case, measurements with one-way communication must be included [38]: imposing no-signaling amounts to admitting all bipartite process functions, hence all causal orders compatible with the no-self-signaling and no-two-way-signaling constraints. More generally, logical consistency of deterministic classical communication is equivalent with the normalization and no-signaling constraints that define the disjunctive FR product.

This yields a new characterization of process functions in terms of no-signaling and normalization conditions, complementing previously established characterizations [24, 25, 26, 32]. The disjunctive FR product can also be used to witness classical non-causality: if a process function corresponds to a hyperedge of the disjunctive but not the maximal FR product (which covers all causal process functions), it is non-causal. Furthermore, the no-signaling conditions of the disjunctive FR product yield substitution rules that allow to generate non-causal process functions from no-communication ones.

In fact, two connections related to our main result were already hinted at in Ref. [34]. First, unit effects in boxworld can be identified with classical processes; in the deterministic case these are process functions, so that process functions correspond to deterministic unit effects. The deterministic wiring and non-wiring measurements of Ref. [34] therefore correspond to causal and non-causal process functions, respectively, and their algorithms generate process functions. Second, the authors of Ref. [34] also noted an equivalence between deterministic unit effects and the hyperedges of the FR product of unipartite Bell scenarios. Lemma 1 proves it explicitly: deterministic unit effects determine exactly the sets of events satisfying Eq. (39), and these are exactly the hyperedges of the disjunctive FR product. Together, these observations suggest the link between hyperedges and process functions that we establish rigorously here, together with the precise relation between logical consistency and the combinatorial structure of Bell scenarios. In a similar vein, Ref. [51] proposed a graph-theoretic criterion for logical consistency, the siblings-on-cycles (SOC) property, which constrains the graph encoding of the causal order of the parties and, through the above equivalence, also identifies the valid unit effects. Since unit effects specify the valid hyperedges of the disjunctive FR product [34], the SOC property equally constrains the admissible hyperedges.

Our results thus add to a growing web of connections between process functions and other objects. The equivalence between the Lugano process, a tripartite non-causal process function, and the SHIFT measurement, which exhibits quantum nonlocality without entanglement (QNLWE) [52], was first pointed out in Ref. [35] and later generalized to arbitrary dimensions in Ref. [26]. This generalization rests on unambiguous product bases, which are in correspondence with all process functions [26], and whose events generate trivial Bell inequalities [43, 44], i.e., inequalities satisfied by all no-signaling correlations. This last property is precisely what we exploit in the proof of Theorem 4, where we show that such inequalities are moreover saturated by every no-signaling model – so that the events in question are exactly those generated by the disjunctive FR product of unipartite Bell scenarios. We thereby obtain the first joint hypergraph representation of non-causality and an event-based notion of NLWE – one that the no-signaling substitution rules make straightforward to construct. Indeed, our result highlights that these phenomena are not specific to quantum theory. This resonates with previous appearances of a SHIFT-type measurement within non-quantum theories, i.e. generalized probabilistic theories (GPTs) [17, 53, 54]: it was first identified in boxworld in Ref. [55] (proof of Theorem 3 and Fig. 2), and later rediscovered in Ref. [34] (Section VII B) where it was explicitly presented as a deterministic non-wiring, which we identify as a Lugano-like process function. It was also shown by Pusey [56] to arise in Spekkens’ toy theory [57], which reproduces many quantum phenomena except nonlocality and contextuality. Our work thus contributes to a recent line of research on indefinite causal order beyond quantum theory [58, 59, 60, 61, 62].

In a different direction, although process functions are usually presented as the classical version of quantum indefinite causal order [19, 20], they have been given some alternative readings. They have been related to agent-environment interactions in artificial intelligence [63], and to a recent interpretation of indefinite causal order in which agents are groupings of input and output labels (“agents from bits” [64]). In line with the latter, we propose the notions of agent and process events, which encode two complementary perspectives: the one of the local agents performing operations and the one of the global superobserver represented by the process itself (see Appendix § 1.1). In the same spirit, and following the relation established with Ref. [34], our results recast process functions – non-causal ones included – as the logically consistent deterministic compositions of black boxes.

Finally, a natural question for future work is whether the causal hierarchy of FR products has observational consequences. For classical, quantum, and boxworld models it does not: the correlations are insensitive to which FR product is used [9, 15], so that the richer communication structures – non-causal process functions in particular – leave no observational trace in this context. Other theories are more discriminating: Q1Q_1 models [9], Prop. 6.3.1 distinguish the minimal product from the rest, yielding almost quantum correlations [18] beyond it. This raises the question whether some alternative theory is not observationally equivalent for the maximal and the disjunctive FR product. In such a theory, including the non-causal measurements in a Bell scenario would change its no-signaling correlations, so that classical non-causality would acquire a new kind of signature which may offer a new handle on it, or an argument for ruling it out.

Acknowledgments

We thank Mirjam Weilenmann for enlightening discussions about the connections between unit effects, classical processes, and hyperedges via e-mail and at Causalworlds 2026. Moreover, we thank Ana Belén Sainz and Elie Wolfe for helpful discussions. We acknowledge the use of the AI assistant Claude Opus 5.5 (Anthropic). Interaction with it helped complete the proof of Theorem 4, in particular through Lemma 1. It was further used to check the proofs of Lemma 2 and Corollary 4. The assistant was finally used to improve the English of the manuscript. All mathematical statements and proofs were independently derived, checked, and are the sole responsibility of the authors. H.D., A.L., and A.A. acknowledge financial support from the Government of Spain (Severo Ochoa CEX2019-000910-S, NextGenerationEU PRTR-C17.I1 and FUNQIP), Fundació Cellex, Fundació Mir-Puig, Generalitat de Catalunya (CERCA program). A.A was also supported by the ERC AdG CERQUTE, the AXA Chair in Quantum Information Science. K.S. and A.L. acknowledge: This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/PAT4559623. For open access purposes, the author has applied a CC BY public copyright license to any author-accepted manuscript version arising from this submission. A.L. acknowledges the European Union (PASQuanS2.1, 101113690). A.L. benefited from the activities of COST Action CA23115: Relativistic Quantum Information, funded by COST (European Cooperation in Science and Technology), specifically discussions during the conference Observers and Causality in Quantum Gravity in Bratislava, Slovakia (2026). R.K. acknowledges: This work received support from the French government under the France 2030 investment plan, as part of the Initiative d’Excellence d’Aix-Marseille Université-A*MIDEX, AMX-22-CEI-01. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Commission. Neither the European Union nor the granting authority can be held responsible for them.

References

Virtual hyperedges

A contextuality scenario HH may allow for additional hyperedges:

Definition 13 (Virtual edges). A probabilistic model on a contextuality scenario HH can allow for virtual hyperedges that are not contained in HH but satisfy the normalization condition (Eq. (3)).

Intuitively, one may consider virtual hyperedges as those sets of vertices that “want to be” a measurement [9]. To illustrate this definition, consider the pentagon scenario shown in Figure 1b whose normalization conditions are

p(v1)+p(v2)=1(I)p(v_1) + p(v_2) = 1 \quad\text{(I)}
p(v2)+p(v3)=1(II)p(v_2) + p(v_3) = 1 \quad\text{(II)}
p(v3)+p(v4)=1(III)p(v_3) + p(v_4) = 1 \quad\text{(III)}
p(v4)+p(v5)=1(IV)p(v_4) + p(v_5) = 1 \quad\text{(IV)}
p(v5)+p(v1)=1(V).p(v_5) + p(v_1) = 1 \quad\text{(V)}.

Subtracting (IV) from (III) yields

p(v3)−p(v5)=0.p(v_3) - p(v_5) = 0.

Adding (V) to this equation gives

p(v3)+p(v1)=1,p(v_3) + p(v_1) = 1,

which shows that the dashed hyperedge in Figure 1b (bottom) is indeed normalized and thus constitutes a virtual hyperedge.

Contextuality scenarios such as the above example are called incomplete, while one may construct a completion of it:

Definition 14 (Completion). The completion H‾\overline{H} of a contextuality scenario HH is obtained by adding all the virtual edges to HH.

This allows to define a multipartite FR product that contains all other ones:

Definition 15 (Def. 10 in [15]). The completion product is defined as the completion of any other multipartite FR product:

⨂i=1nHi=⨂i=1nHi‾.\bigotimes_{i=1}^{n} H_i = \overline{\bigotimes_{i=1}^{n} H_i}.

This product is in general larger than the disjunctive FR product. In this work, we consider unipartite Bell scenarios, for which the equivalence between deterministic unit effects and hyperedges of the disjunctive FR product of Bell scenarios forbids the existence of virtual hyperedges (addressing Conjecture 2 of [15] for the case of unipartite Bell scenarios). In fact, this follows immediately from Lemma 1:

Corollary 4. The disjunctive FR product of Bell scenarios has no virtual hyperedges; that is, it coincides with the completion FR product.

Proof. Assume a virtual hyperedge eVe_{\mathrm{V}} exists. In general, its vertices may not satisfy pairwise exclusivity. Its normalization condition is implied by the normalization conditions of the hyperedges of disj⁡⨂k=1nB1,mk,dk\operatorname{disj}\bigotimes_{k=1}^{n} B_{1,m_k,d_k} and thus ∑v∈eVp(v)=1\sum_{v \in e_{\mathrm{V}}} p(v) = 1 holds for any pp defined over disj⁡⨂k=1nB1,mk,dk\operatorname{disj}\bigotimes_{k=1}^{n} B_{1,m_k,d_k}. By the first direction of Lemma 1 this identity holds for any no-signaling model, i.e., Eq. (39) is satisfied by eVe_{\mathrm{V}}. By the reverse direction of Lemma 1, any set of vertices satisfying Eq. (39) is already a hyperedge of disj⁡⨂k=1nB1,mk,dk\operatorname{disj}\bigotimes_{k=1}^{n} B_{1,m_k,d_k}. Thus eVe_{\mathrm{V}} is not a virtual hyperedge.

Matrix representation of the disjunctive Foulis-Randall product

Following [15], one can formalize the requirements of normalization and no-signaling using matrix representations of hypergraphs.

Definition 16 (Matrix representation of a hypergraph). The matrix MM of a hypergraph HH has rows that are labeled by the hyperedges e∈E(H)e \in E(H) and columns that are labeled by the vertices v∈V(H)v \in V(H), i.e., MevM_{e v} corresponds to a potential outcome vv of measurement ee. If outcome vv appears in ee, Mev=1M_{e v} = 1, and if not, Mev=0M_{e v} = 0.

Normalization of probabilistic models can be expressed the following way:

Definition 17 (Matrix representation of a normalization condition). The normalization condition for a probabilistic model pp defined over a hypergraph HH can be written as M⋅p=1M \cdot p = 1 where the components of pp are the probabilities p(v),v∈V(H)p(v), v \in V(H) and [1]i=1 ∀i=1,…,∣E(H)∣[1]_i = 1 \ \forall i = 1, \ldots, |E(H)|.

To enforce that the hyperedges defining MM satisfy all no-signaling and normalization conditions, the rank of the hypergraph matrix MM has to be equal to the rank of another matrix whose rows are defined by no-signaling and normalization conditions. The latter two are defined in the following.

Definition 18 (Composite normalization matrix). Given a composite contextuality scenario H=⨂i=1nHiH = \bigotimes_{i=1}^{n} H_i, the composite normalization matrix MNormM_{\mathrm{Norm}} has {0,1}\{0,1\}-valued entries, its columns are labeled by the vertices and its rows correspond to no-communication hyperedges: these correspond to each scenario kk choosing a local measurement ek∈Hke_k \in H_k independently of the other scenarios. MNormM_{\mathrm{Norm}} represents all possible combinations of such local choices and satisfies

MNorm⋅p=1.(57)M_{\mathrm{Norm}} \cdot p = 1. \tag*{(57)}

One may rewrite the no-signaling conditions as

∑vk∈ekp(v1,…,vk,…,vn)−∑vk∈ek′p(v1,…,vk,…,vn)=0(58)\sum_{v_k\in e_k} p(v_1,\ldots,v_k,\ldots,v_n)-\sum_{v_k\in e'_k}p(v_1,\ldots,v_k,\ldots,v_n)=0 \tag*{(58)}
∀ek,ek′∈Hk, ∀v1,…,vk−1,vk+1,…,vn,\forall e_k,e'_k\in H_k,\ \forall v_1,\ldots,v_{k-1},v_{k+1},\ldots,v_n,

which allows to define the following matrix:

Definition 19 (Composite no-signaling matrix). We denote MNSM_{\mathrm{NS}} as the matrix which represents all possible no-signaling equalities as defined in Eq. (58). The rows of MNSM_{\mathrm{NS}} correspond to a specific no-signaling condition, while the columns are labelled by composite events v:=(v1,…,vn)\mathbf{v}:=(v_1,\ldots,v_n). A value of +1+1 is assigned to a matrix element if the column label contains vk∈ek\ek′v_k\in e_k\backslash e'_k, −1-1 if vk∈ek′\ekv_k\in e'_k\backslash e_k and 00 for all other cases. Thus, Eq. (58) can be written as

MNS⋅p=0(59)M_{\mathrm{NS}}\cdot p=0 \tag*{(59)}

where [0]i=0 ∀i=1,…,∣E(H)∣[\mathbf{0}]_i=0\ \forall i=1,\ldots,|E(H)|.

Given these definitions, HH is said to capture no-signaling and normalization if and only if the reduced row echelon form (rref) of the hypergraph matrix M1M^1 (padded with a column of all-ones) equals the reduced row echelon form of the combined rows of normalization and no-signaling hyperedges, MNorm1∪MNS0M^1_{\mathrm{Norm}}\cup M^0_{\mathrm{NS}} (padded with a column of all-ones and all-zeros, respectively):

rref⁡(M1)=rref⁡(MNorm1∪MNS0)(60)\operatorname{rref}(M^1)=\operatorname{rref}(M^1_{\mathrm{Norm}}\cup M^0_{\mathrm{NS}}) \tag*{(60)}

We can use these definitions to provide a matrix-based definition of the disjunctive FR product disj⊗k=1nHk\mathop{\mathrm{disj}\otimes}_{k=1}^{n}H_k for nn local scenarios HkH_k. As discussed in the main text, one constructs a composite, faithful orthogonality graph from the orthogonality graphs Ortho⁡(Hk)\operatorname{Ortho}(H_k) (cf. Def. 10). Then one inspects all of its cliques, i.e., sets of events that are mutually connected. These events satisfy pairwise exclusivity. Considering these as candidate hyperedges, the sum of probabilities over their respective events must be one. If these conditions do not introduce anything beyond the no-signaling and normalization constraints, then the candidate hyperedge is included in the edge set E(disj⊗k=1nHk)E(\mathop{\mathrm{disj}\otimes}_{k=1}^{n}H_k). Mathematically, this can be expressed by comparing the ranks of two suitably defined matrices.

Definition 20 (Disjunctive FR product). The disjunctive FR product disj⊗k=1nHk\mathop{\mathrm{disj}\otimes}_{k=1}^{n}H_k is defined by vertices V(disj⊗k=1nHk):=×k=1nV(Hk)V(\mathop{\mathrm{disj}\otimes}_{k=1}^{n}H_k):=\mathop{\times}_{k=1}^{n}V(H_k) and the edge set

E(disj⊗i=1nHi):={ee∈Cliques⁡(FaithfulOrthoGraph⁡(H1,…,Hn))∧ rank⁡(MNorm1∪MNS0)=rank⁡(e1∪MNorm1∪MNS0)}.(61)E(\mathop{\mathrm{disj}\otimes}_{i=1}^{n}H_i):= \left\{ \begin{array}{ll} e & e\in\operatorname{Cliques}(\operatorname{FaithfulOrthoGraph}(H_1,\ldots,H_n))\\ &\land\ \operatorname{rank}(M^1_{\mathrm{Norm}}\cup M^0_{\mathrm{NS}}) =\operatorname{rank}(e^1\cup M^1_{\mathrm{Norm}}\cup M^0_{\mathrm{NS}}) \end{array} \right\}. \tag*{(61)}

where Cliques⁡\operatorname{Cliques} denotes subsets of vertices that are adjacent according to the faithful orthogonality graph FaithfulOrthoGraph, while MNorm1M^1_{\mathrm{Norm}} and MNS0M^0_{\mathrm{NS}} denote the matrix representation of normalization and no-signaling conditions (Def. 18 and Def. 19) padded with a column of ones or rather zeroes. e1e^1 is a short notation for the matrix representation of ∑v∈ep(v)=1\sum_{\mathbf{v}\in e}p(\mathbf{v})=1.

One can also reformulate the characterization of Corollary 2:

Corollary 5. A no-self-signaling function w:A→Xw:\mathcal{A}\to\mathcal{X} is a process function if and only if it gives rise to events e:={(a∣w(a))}e:=\{(a|w(a))\} that satisfy

rank⁡(MNorm1∪MNS0)=rank⁡(e1∪MNorm1∪MNS0)(62)\operatorname{rank}(M^1_{\mathrm{Norm}}\cup M^0_{\mathrm{NS}}) =\operatorname{rank}(e^1\cup M^1_{\mathrm{Norm}}\cup M^0_{\mathrm{NS}}) \tag*{(62)}

where MNorm1M^1_{\mathrm{Norm}} and MNS0M^0_{\mathrm{NS}} are the matrix form of the no-signaling and normalization conditions specified in Def. 18 and Def. 19. e1e^1 denotes the matrix form of the normalization condition ∑v∈ep(v)\sum_{\mathbf{v}\in e}p(\mathbf{v}) for a probabilistic model pp.

Finally, one may also state the definition of virtual hyperedges (cf. Appendix A) using matrices:

Definition 21. A hyperedge ee is a virtual hyperedge of HH if and only if

e∉E(H)∧rank⁡(M1(H)∪e1)=rank⁡(M1(H))(63)e\notin E(H)\land\operatorname{rank}(M^1(H)\cup e^1)=\operatorname{rank}(M^1(H)) \tag*{(63)}

Comparison with the QC-QC hierarchy

By Theorem 4 of [65], any process function w:A→Xw:\mathcal{A}\to\mathcal{X} admits a unique encoding as a process matrix:

WwA=∑a⨂k∣ak⟩⟨ak∣AkO⊗wk(ak)⟩⟨wk(ak)∣AkI.(64)W_w^{A}=\sum_a\bigotimes_k |a_k\rangle\langle a_k|^{A_k^O}\otimes w_k(a_k)\rangle\langle w_k(a_k)|^{A_k^I}. \tag*{(64)}

with parties input and output spaces of linear operators L(HAkI)\mathcal{L}(\mathcal{H}^{A_k^I}) and L(HAkO)\mathcal{L}(\mathcal{H}^{A_k^O}).

Quantum circuits with operations used in parallel (QC-PAR)

A QC-PAR [33] is a circuit composed of two internal Completely-Positive-Trace-Preserving (CPTP) operations: MP:L(HP)→L(HAIα)\mathcal{M}_{P}:\mathcal{L}(\mathcal{H}^{P})\to\mathcal{L}(\mathcal{H}^{A^I\alpha}), HAIα:=HAI⊗Hα\mathcal{H}^{A^I\alpha}:=\mathcal{H}^{A^I}\otimes\mathcal{H}^{\alpha}, HAI:=⨂k∈NHAkI\mathcal{H}^{A^I}:=\bigotimes_{k\in\mathcal{N}}\mathcal{H}^{A_k^I} (with Choi matrix MP∈L(HPAIα)M_P\in\mathcal{L}(\mathcal{H}^{PA^I\alpha})) which takes the circuit’s inputs in the global past HP\mathcal{H}^{P} and outputs a state in the input Hilbert space HAI\mathcal{H}^{A^I}, which in general may be correlated with an auxiliary system in some Hilbert space Hα\mathcal{H}^{\alpha}; and MF:L(HAOα)→L(HF)\mathcal{M}_{F}:\mathcal{L}(\mathcal{H}^{A^O\alpha})\to\mathcal{L}(\mathcal{H}^{F}) (with Choi matrix MF∈L(HAOαF)M_F\in\mathcal{L}(\mathcal{H}^{A^O\alpha F})) which takes the output state of HAO\mathcal{H}^{A^O} together with the auxiliary state to the circuit’s outputs in

the global future HF\mathcal{H}^{F}. The corresponding process matrix is

W≡MP∗MF∈L(HPANIOαF),(65)W \equiv M_{P} * M_{F} \in\mathcal{L}\left(\mathcal{H}^{P A_{N}^{IO}\alpha_{F}}\right), \tag*{(65)}

where ∗* is the link product [46, 66]. For two operators M∈L(HXY)M \in\mathcal{L}(\mathcal{H}^{XY}) and N∈L(HYZ)N \in\mathcal{L}(\mathcal{H}^{YZ}), the latter is defined as

M∗N=Tr⁡Y[(MTY⊗1Z)(1X⊗N)]M * N = \operatorname{Tr}_{Y}\left[\left(M^{T_{Y}} \otimes\mathbb{1}^{Z}\right)\left(\mathbb{1}^{X} \otimes N\right)\right]

where TYT_{Y} denotes partial transposition on YY, with the operations satisfying the constraints

Tr⁡ANIαMP=1P,(66)\operatorname{Tr}_{A_{N}^{I}\alpha} M_{P} = \mathbb{1}^{P}, \tag*{(66)}

and

Tr⁡F(MP∗MF)=Tr⁡αMP⊗1ANO.(67)\operatorname{Tr}_{F}(M_{P} * M_{F}) = \operatorname{Tr}_{\alpha} M_{P} \otimes\mathbb{1}^{A_{N}^{O}}. \tag*{(67)}

Considering trivial global input and output spaces – alternatively, one could also purify the process matrix Eq. (64) following Ref. [67] – and a trivial auxiliary state, dim⁡(HP)=dim⁡(HF)=dim⁡(Hα)=1\dim(\mathcal{H}^{P}) = \dim(\mathcal{H}^{F}) = \dim(\mathcal{H}^{\alpha}) = 1, the constraints Eq. (C4) becomes

Tr⁡ANIMP=1,(68)\operatorname{Tr}_{A_{N}^{I}} M_{P} = 1, \tag*{(68)}

and

MP∗MF=MP⊗1ANO.(69)M_{P} * M_{F} = M_{P} \otimes\mathbb{1}^{A_{N}^{O}}. \tag*{(69)}

Theorem 5. Process matrices of “all-for-one” process functions generated by the minimal FR product are QC-PARs.

Proof. Relabel the global future party kk of the “all-for-one” process functions as party “FF”: for all k∈N∖Fk \in\mathcal{N} \setminus F, the input state corresponds to constant functions. Tracing out the future system thus leads to a N−1N-1 process matrix of the form Eq. (64). Because the parties’ outputs are linearly independent, averaging over all of them leads to:

Tr⁡FWwA=⨂k1AkO⊗∣xk⟩⟨xk∣AkI,(70)\operatorname{Tr}_{F} W_{w}^{A} = \bigotimes_{k} \mathbb{1}^{A_{k}^{O}} \otimes\lvert x_{k}\rangle\langle x_{k}\rvert^{A_{k}^{I}}, \tag*{(70)}

satisfying Eq. (69) with $M_{P} = \bigotimes_{k} \lvert x_{k}\rangle\langle x_{k}\rvert^{A_{k}^{I}}$. The latter straightforwardly satisfies Eq. (C5)`.

The process

x1:=0,x2:=0,x3:=a1a2,(71)x_{1} := 0,\quad x_{2} := 0,\quad x_{3} := a_{1}a_{2}, \tag*{(71)}

is an example of a process function generate by the minimal FR product, thus implementable in a QC-PAR. Note that this is only true because of the relabeling. If the last party is not relabeled as FF, then “all-for-one” process functions are a special case of processes with fixed causal order in which all but one party are in parallel. Consequently, the minimal FR product does not correspond exactly to QC-PAR, but is slightly more general: the product associated with QC-PAR in the absence of relabelling is the Cartesian product, whose hyperedges involve no communication at all. Since no analogous relabelling is used for the other classes, this should be kept in mind when comparing the two hierarchies.

Quantum circuits with fixed causal order (QC-FO)

A QC-FO [33], also known as quantum comb [46, 47], is a circuit composed of N+1N+1 internal CPTP operations placed in a fixed order (1,…,N)(1,\ldots,N). Considering a trivial global input space, and denoting the reduced matrices

W(n):=1∏k=n+1NdkOTr⁡∏k=n+1NAkIOAFOW,(72)W_{(n)} := \frac{1}{\prod_{k=n+1}^{N} d_{k}^{O}} \operatorname{Tr}_{\prod_{k=n+1}^{N} A_{k}^{IO} A_{F}^{O}} W, \tag*{(72)}

their process matrices are characterized by the following constraints:

Tr⁡FW=W(N)⊗1ANO,(73)\operatorname{Tr}_{F} W = W_{(N)} \otimes\mathbb{1}^{A_{N}^{O}}, \tag*{(73)}
∀n∈N∖N,Tr⁡An+1IW(n+1)=W(n)⊗1AnO,(74)\forall n \in\mathcal{N} \setminus N,\quad\operatorname{Tr}_{A_{n+1}^{I}} W_{(n+1)} = W_{(n)} \otimes\mathbb{1}^{A_{n}^{O}}, \tag*{(74)}
Tr⁡A1IW(1)=1.(75)\operatorname{Tr}_{A_{1}^{I}} W_{(1)} = 1. \tag*{(75)}

Theorem 6. Process matrices of process functions generated by the common FR product are QC-FOs.

Proof. Label the parties according to the fixed order A1≺⋯≺ANA_{1} \prec\cdots\prec A_{N}. Then x1=c1x_{1}=c_{1}, xn=wn(a1,…,an−1)x_{n}=w_{n}(a_{1},\ldots,a_{n-1}) for (n≥2)(n \ge2). Define

W(n):=∑a1,…,an−1⨂j=1n−1(∣aj⟩⟨aj∣AjO⊗∣xj⟩⟨xj∣AjI)⊗∣wn⟩⟨wn∣AnI,(76)W_{(n)} := \sum_{a_{1},\ldots,a_{n-1}} \bigotimes_{j=1}^{n-1}\left(\lvert a_{j}\rangle\langle a_{j}\rvert^{A_{j}^{O}} \otimes\lvert x_{j}\rangle\langle x_{j}\rvert^{A_{j}^{I}}\right) \otimes\lvert w_{n}\rangle\langle w_{n}\rvert^{A_{n}^{I}}, \tag*{(76)}

where xj=wj(a1,…,aj−1)x_{j}=w_{j}(a_{1},\ldots,a_{j-1}). Since W(1)=∣c1⟩⟨c1∣A1IW_{(1)}=\lvert c_{1}\rangle\langle c_{1}\rvert^{A_{1}^{I}}, we have Tr⁡A1IW(1)=1\operatorname{Tr}_{A_{1}^{I}}W_{(1)}=1 (Eq. (75)). Moreover, for every n=1,…,N−1n=1,\ldots,N-1,

Tr⁡An+1IW(n+1)=W(n)⊗∑an∣an⟩⟨an∣AnO=W(n)⊗1AnO,(77)\operatorname{Tr}_{A_{n+1}^{I}}W_{(n+1)}=W_{(n)}\otimes\sum_{a_{n}}\lvert a_{n}\rangle\langle a_{n}\rvert^{A_{n}^{O}}=W_{(n)}\otimes\mathbb{1}^{A_{n}^{O}}, \tag*{(77)}

thus Eq. (C11) is satisfied. Finally, the last output cannot influence any input, and hence

WwA=W(N)⊗∑aN∣aN⟩⟨aN∣ANO=W(N)⊗1ANO.W_{w}^{A}=W_{(N)}\otimes\sum_{a_{N}}\lvert a_{N}\rangle\langle a_{N}\rvert^{A_{N}^{O}}=W_{(N)}\otimes\mathbb{1}^{A_{N}^{O}}.

Taking FF to be trivial gives Tr⁡FWwA=W(N)⊗1ANO\operatorname{Tr}_{F}W_{w}^{A}=W_{(N)}\otimes\mathbb{1}^{A_{N}^{O}} (Eq. (73)). Thus all the QC-FO constraints are satisfied.

The process

x1:=0,x2:=a1,x3:=a1a2,(78)x_{1}:=0,\quad x_{2}:=a_{1},\quad x_{3}:=a_{1}a_{2}, \tag*{(78)}

is an example of a process function generate by the common FR product, thus implementable in a QC-FO. It is straightforward to see that its process matrix does not satisfy Eq. (68), and thus it cannot be implemented in a QC-PAR.

Quantum circuits with classical control of causal order (QC-CC)

A QC-CC [33], is a circuit composed of N+1N+1 internal quantum instruments (collections of CP maps which sum up to a CPTP map) in which the causal order of external operations can be established dynamically, with the operations in the past determining the causal order of the operations in the future. Their process matrices are characterized by the following constraints, for all nonempty ordered subsets (k1,…,kn)(k_1,\ldots,k_n) of N\mathcal{N}:

W=∑(k1,…,kN)W(k1,…,kN,F),(79)W=\sum_{(k_1,\ldots,k_N)}W_{(k_1,\ldots,k_N,F)}, \tag*{(79)}
∀(k1,…,kN),Tr⁡FW(k1,…,kN,F)=W(k1,…,kN)⊗1ANO,(80)\forall(k_1,\ldots,k_N),\quad\operatorname{Tr}_{F}W_{(k_1,\ldots,k_N,F)}=W_{(k_1,\ldots,k_N)}\otimes\mathbb{1}^{A_N^O}, \tag*{(80)}
∀n=1,…,N−1,∀(k1,…,kn),∑kn+1Tr⁡Akn+1IW(k1,…,kn+1)=W(k1,…,kn)⊗1AnO,(81)\forall n=1,\ldots,N-1,\quad\forall(k_1,\ldots,k_n),\quad\sum_{k_{n+1}}\operatorname{Tr}_{A_{k_{n+1}}^I}W_{(k_1,\ldots,k_{n+1})}=W_{(k_1,\ldots,k_n)}\otimes\mathbb{1}^{A_n^O}, \tag*{(81)}
∑k1Tr⁡Ak1IW(k1)=1.(82)\sum_{k_1}\operatorname{Tr}_{A_{k_1}^I}W_{(k_1)}=1. \tag*{(82)}

Theorem 7. Process matrices of process functions generated by the maximal FR product are QC-CCs.

Proof. A measurement protocol of the maximal FR product is a rooted decision tree. At every node, the protocol selects a party that has not acted before, and the outcome of that party determines the next node. Although this representation selects one party at each step, several parties may be causally parallel. In particular, several parties may receive constant inputs and therefore be causally initial. Such parallel parties can be placed in any order without introducing a causal dependence between them.

We fix a sequentialization of the protocol. Thus, every output string aa determines an ordered branch

π(a)=(k1,…,kN)(83)\pi(a)=(k_1,\ldots,k_N) \tag*{(83)}

of the parties.

Here, k1k_1 is only the first party in the chosen sequentialization; it need not be the unique causally initial party. If kjk_j is parallel to the parties preceding it, then its input is independent of their outcomes. Thus, along a branch (k1,…,kN)(k_1,\ldots,k_N), we may write

wkn(a\kn)=wkn(ak1,…,akn−1).(84)w_{k_n}(a_{\backslash k_n})=w_{k_n}(a_{k_1},\ldots,a_{k_{n-1}}). \tag*{(84)}

This follows directly from the recursive definition of a measurement protocol: once a party has been selected, it cannot be selected again.

For an ordered subset (k1,…,kn)(k_1,\ldots,k_n), (ak1,…,akn−1)(a_{k_1},\ldots,a_{k_{n-1}}) is the set of outcome histories for which the chosen sequentialization selects the parties k1,…,knk_1,\ldots,k_n in this order. Define

W(k1,…,kn):=∑(ak1,…,akn−1)⨂j=1n−1(∣akj⟩⟨akj∣⊗∣wkj(ak1,…,akj−1)⟩⟨wkj(ak1,…,akj−1)∣)AkjOAkjI⊗∣wkn(ak1,…,akn−1)⟩⟨wkn(ak1,…,akn−1)∣AknI.(85)W_{(k_1,\ldots,k_n)}:=\sum_{(a_{k_1},\ldots,a_{k_{n-1}})}\bigotimes_{j=1}^{n-1}\left(|a_{k_j}\rangle\langle a_{k_j}|\otimes|w_{k_j}(a_{k_1},\ldots,a_{k_{j-1}})\rangle\langle w_{k_j}(a_{k_1},\ldots,a_{k_{j-1}})|\right)_{A_{k_j}^O A_{k_j}^I} \otimes|w_{k_n}(a_{k_1},\ldots,a_{k_{n-1}})\rangle\langle w_{k_n}(a_{k_1},\ldots,a_{k_{n-1}})|_{A_{k_n}^I}. \tag*{(85)}

Here and below, an operator is defined to be zero if the corresponding ordered subset does not occur in the chosen sequentialization.

Let k1k_1 be the formal first party in this sequentialization. Since k1k_1 is causally initial, its input is a constant ck1c_{k_1}, and hence W(k1)=∣ck1⟩⟨ck1∣Ak1IW_{(k_1)}=|c_{k_1}\rangle\langle c_{k_1}|_{A_{k_1}^I}. Therefore,

∑k1Tr⁡Ak1IW(k1)=Tr⁡(∣ck1⟩⟨ck1∣Ak1I)=1,(86)\sum_{k_1}\operatorname{Tr}_{A_{k_1}^I}W_{(k_1)}=\operatorname{Tr}\left(|c_{k_1}\rangle\langle c_{k_1}|_{A_{k_1}^I}\right)=1, \tag*{(86)}

which proves Eq. (C19).

Consider now a prefix (k1,…,kn)(k_1,\ldots,k_n) with n<Nn<N. For each compatible history (ak1,…,akn−1)(a_{k_1},\ldots,a_{k_{n-1}}), the outcome akna_{k_n} determines exactly one next party kn+1k_{n+1}. Thus, as kn+1k_{n+1} ranges over the parties not yet selected, the operators W(k1,…,kn+1)W_{(k_1,\ldots,k_{n+1})} partition all possible extensions of the prefix by the outcome akna_{k_n}. Moreover, tracing out the input of the next party removes its normalized input state, since

Tr⁡(∣wkn+1(ak1,…,akn)⟩⟨wkn+1(ak1,…,akn)∣Akn+1I)=1.(87)\operatorname{Tr}\left(|w_{k_{n+1}}(a_{k_1},\ldots,a_{k_n})\rangle\langle w_{k_{n+1}}(a_{k_1},\ldots,a_{k_n})|_{A_{k_{n+1}}^I}\right)=1. \tag*{(87)}

It follows that

∑kn+1Tr⁡Akn+1IW(k1,…,kn,kn+1)=W(k1,…,kn)⊗∑akn∣akn⟩⟨akn∣AknO=W(k1,…,kn)⊗1AknO.(88)\begin{aligned} \sum_{k_{n+1}}\operatorname{Tr}_{A_{k_{n+1}}^I}W_{(k_1,\ldots,k_n,k_{n+1})} &=W_{(k_1,\ldots,k_n)}\otimes\sum_{a_{k_n}}|a_{k_n}\rangle\langle a_{k_n}|_{A_{k_n}^O}\\ &=W_{(k_1,\ldots,k_n)}\otimes\mathbb{1}_{A_{k_n}^O}. \tag*{(88)} \end{aligned}

The first equality follows because every compatible pair (ak1,…,akn−1,akn)(a_{k_1},\ldots,a_{k_{n-1}},a_{k_n}) occurs exactly once in the sum, namely in the term corresponding to the next party selected by the protocol. The second equality follows from the completeness of the classical output basis. This is precisely the recursive QC-CC constraint in Eq. (C18).

For a complete branch (k1,…,kN)(k_1,\ldots,k_N), take the final system FF to be one-dimensional and define

W(k1,…,kN,F):=W(k1,…,kN)⊗1AkNO⊗∣0⟩⟨0∣F.(89)W_{(k_1,\ldots,k_N,F)}:=W_{(k_1,\ldots,k_N)}\otimes\mathbb{1}_{A_{k_N}^O}\otimes|0\rangle\langle0|^F. \tag*{(89)}

It then follows that Eq. (C17) is satisfied. Finally, every output string aa follows exactly one complete branch of the protocol. Expanding the identity on the output of the last party, 1AkNO=∑akN∣akN⟩⟨akN∣\mathbb{1}^{A^O_{k_N}} = \sum_{a_{k_N}} |a_{k_N}\rangle\langle a_{k_N}|, and summing over all complete branches therefore gives

∑(k1,…,kN)W(k1,…,kN,F)=WwA⊗∣0⟩⟨0∣F,(90)\sum_{(k_1,\ldots,k_N)} W_{(k_1,\ldots,k_N,F)} = W^A_w \otimes|0\rangle\langle0|^F, \tag*{(90)}

which is Eq. (79).

All the operators constructed above are positive semidefinite and satisfy Eqs. (79)–(82). Hence WwAW^A_w is a QC-CC process matrix.

Corollary 6. Every classical deterministic QC-CC process that is dynamical, i.e. not QC-FO, has influenceable classical control of causal order.

Proof. By Theorem 7, every process matrix of a maximal FR process function is a QC-CC. Let us define two subclasses of QC-CC introduced in Ref. [45]. QC-convFO is the convex hull of fixed-order circuits, W=∑πqπWπW = \sum_\pi q_\pi W_\pi, with qπ≥0q_\pi\ge0, ∑πqπ=1\sum_\pi q_\pi= 1, where each WπW_\pi is compatible with the fixed order π\pi. QC-NICC is the class of QC-CC processes admitting a realization in which the classical control of the causal order cannot be influenced by the parties’ CPTP operations. In general,

QC-FO⊆QC-convFO⊆QC-NICC⊆QC-CC.(91)\mathrm{QC\text{-}FO} \subseteq\mathrm{QC\text{-}convFO} \subseteq\mathrm{QC\text{-}NICC} \subseteq\mathrm{QC\text{-}CC}. \tag*{(91)}

It therefore suffices to show that any process belonging to QC-NICC is in QC-convFO. This is straightforward, as for deterministic classical process matrices:

QC-convFOdet.cl.=QC-NICCdet.cl.=QC-FOdet.cl..(92)\mathrm{QC\text{-}convFO}^{\mathrm{det.cl.}} = \mathrm{QC\text{-}NICC}^{\mathrm{det.cl.}} = \mathrm{QC\text{-}FO}^{\mathrm{det.cl.}}. \tag*{(92)}

Indeed, a deterministic classical process function assigns a unique input string x=w(a)x = w(a) to every output string aa. If it admits a convex decomposition Ww=∑λqλWλW_w = \sum_\lambda q_\lambda W_\lambda, qλ>0q_\lambda> 0, then each WλW_\lambda must produce the same assignment x=w(a)x = w(a) for every aa; otherwise the mixture would assign nonzero probability to a different input string and would not be deterministic. Hence every component with nonzero weight realizes the same process function. It follows that, in a QC-convFO decomposition, each fixed-order component with nonzero weight realizes ww, so ww is itself compatible with that fixed order; QC-convFOdet.cl.=QC-FOdet.cl.\mathrm{QC\text{-}convFO}^{\mathrm{det.cl.}} = \mathrm{QC\text{-}FO}^{\mathrm{det.cl.}}. Likewise, in a QC-NICC realization, the distribution of orders is independent of the parties’ operations, and determinism forces every order occurring with nonzero probability to realize the same function ww. Since each complete branch has a fixed causal order, ww must be compatible with every order occurring with nonzero probability, and therefore with at least one fixed order. Hence QC-NICCdet.cl.⊆QC-convFOdet.cl.\mathrm{QC\text{-}NICC}^{\mathrm{det.cl.}} \subseteq\mathrm{QC\text{-}convFO}^{\mathrm{det.cl.}}. The converse follows from the general inclusion QC-convFO ⊆\subseteq QC-NICC.

Agency in process functions

It has as recently been highlighted by Grinbaum [64] that bridging indefinite causal order in the operational approach with dynamical theories in spacetime requires to define agents (or observers) within the theory and in a perspectival way. In particular, the global process is treated as prior, while agents are “admissible groupings of input and output data in a global Hilbert space” (“Agents from bits”). The original data is unlabeled—it is only by dividing the data into inputs and outputs and grouping it by assigning labels (e.g., “Alice” or “Bob”) that observers arise. Redefining these groupings may even eliminate non-causality. In particular, agency is treated as relative to the operational perspectives that result from the definition of agency. The grouping, however is not arbitrary: the process needs to satisfy logical consistency that rules out paradoxes, therefore restricting the possible groupings. This philosophical stance on agency is referred to as operational eternalism, in analogy to the eternalist, “block-universe” view, in which all spacetime events are treated on the same footing.

Inspired by this notion of agency, we inspect the process function scenario. In particular, we can identify two types of events. As depicted in Figure 12, aa and xx can play different roles: from the perspective of the process, aa are inputs and xx outputs, defining a set of process events (x∣a)(x|a) with x=w(a)x = w(a). From the perspective of the parties, xx are inputs (or rather, measurement settings) and aa are outputs (or rather, measurement outcomes), defining agents events (a∣x)(a|x).

Figure 12. Interpretation of agency in process functions. Compared to Figure 4 and Figure 5 we employ a coarse-grained view of the retrocausal identity channels and parties: nn isolated parties are placed in the right blue box, where each party kk receives a classical variable xkx_k (defining the global input x:=(x1,…,xn)\boldsymbol{x} := (x_1,\ldots,x_n)) to which they can apply an arbitrary local operation fkf_k (defining a global operation f:=(f1,…,fn)f := (f_1,\ldots,f_n)) and producing an output aka_k (defining a global output a:=(a1,…,an)a := (a_1,\ldots,a_n)). aa serves as input to the process function ww, which returns xx as output. The events from the perspectives of the process function and the local parties, which we denote process events and agents events, respectively, are related by a simple swap of aa and xx (see top arrow). There is a deeper mathematical representation of this connection (see bottom arrow): the process events define a process function that satisfy logical consistency (cf. Def. 6) [23], while the agents events that satisfy pairwise-exclusive completeness (Theorem 2) [32] and define the outcomes of a no-signaling measurement. This work provides a hypergraph representation of this equivalence.

Thus process and agents events are ultimately the same events, but they are structured differently: the former globally, from the perspective of the process, the latter locally, from that of a single agent. On the one hand, process events are uniquely determined by aa alone, i.e., one can identify them in a coarse-grained, setting-independent manner. On the other hand, identifying the local events (ak∣xk)(a_k|x_k) of an agent requires both outcomes and settings, i.e., one can identify local events only in a fine-grained, setting-dependent manner. One may then associate agency with respect to these two different perspectives: process events correspond to a superobserver that is trivial in the sense that its outputs xx can be identified from aa alone. This global perspective is described as “observer-indefinite” in Ref. [64]. The trivial events of this global superobserver are themselves the coarse-graining of the local events that can be associated with local agents.

The equivalence between these two perspectives, related by a simple swap operation (Figure 12), is grounded by the operational equivalence of their validity: process events define a function that is logically consistent, while agents events are complete and satisfy pairwise exclusivity. Our work provides a hypergraph representation of this equivalence.

References

  1. [1]J. S. Bell, On the Einstein Podolsky Rosen Paradox, Physics Physique Fizika 1, 195 (1964).
  2. [2]J. S. Bell, On the Problem of Hidden Variables in Quantum Mechanics, Reviews of Modern Physics 38, 447 (1966).
  3. [3]N. Brunner, D. Cavalcanti, S. Pironio, V. Scarani, and S. Wehner, Bell nonlocality, Reviews of Modern Physics 86, 419 (2014).
  4. [4]S. Kochen and E. P. Specker, The Problem of Hidden Variables in Quantum Mechanics, Journal of Mathematics and Mechanics 17, 59 (1967).
  5. [5]C. Budroni, A. Cabello, O. Gü hne, M. Kleinmann, and J.-Å. Larsson, Kochen-specker contextuality, Reviews of Modern Physics 94, 045007 (2022).DOI
  6. [6]R. W. Spekkens, Contextuality for preparations, transformations, and unsharp measurements, Physical Review A 71, 052108 (2005).
  7. [7]A. Cabello, S. Severini, and A. Winter, Graph-theoretic approach to quantum correlations, Physical Review Letters 112, 040401 (2014).DOI
  8. [8]S. Abramsky and A. Brandenburger, The sheaf-theoretic structure of non-locality and contextuality, New Journal of Physics 13, 013036 (2011).
  9. [9]A. Acín, T. Fritz, A. Leverrier, and A. B. Sainz, A combinatorial approach to nonlocality and contextuality, Communications in Mathematical Physics 334, 533 (2015).DOI
  10. [10]A. Wilce, Test spaces, Handbook of Quantum Logic and Quantum Structures , 443 (2009).DOI
  11. [11]B. Coecke, D. Moore, and A. Wilce, Operational quantum logic: An overview, Current Research in Operational Quantum Logic: Algebras, Categories, Languages , 1 (2000).
  12. [12]A. Cabello, S. Severini, and A. Winter, (Non-)Contextuality of Physical Theories as an Axiom (2010), arXiv:1010.2163 [quant-ph].arxiv.org/abs/1010.2163
  13. [13]R. Ramanathan, A. Soeda, P. Kurzyński, and D. Kaszlikowski, Generalized Monogamy of Contextual Inequalities from the No-Disturbance Principle, Physical Review Letters 109, 10.1103/PhysRevLett.109.050404 (2012).DOI
  14. [14]D. J. Foulis and C. H. Randall, Empirical logic and tensor products, in Interpretations and foundations of quantum theory, Vol. 5 (1981) pp. 9–20.
  15. [15]A. B. Sainz and E. Wolfe, Multipartite composition of contextuality scenarios, Foundations of Physics 48, 925 (2018).DOI
  16. [16]S. Popescu and D. Rohrlich, Quantum nonlocality as an axiom, Foundations of Physics 24, 379 (1994).DOI
  17. [17]J. Barrett, Information processing in generalized probabilistic theories, Physical Review A 75, 032304 (2007).DOI
  18. [18]M. Navascués, Y. Guryanova, M. J. Hoban, and A. Acín, Almost quantum correlations, Nature Communications 6, 6288 (2015).DOI
  19. [19]F. Costa, G. Rubino, C. Branciard, Časlav Brukner, and M. T. Quintino, Indefinite quantum causality (2026), arXiv:2606.19438 [quant-ph].arxiv.org/abs/2606.19438
  20. [20]O. Oreshkov, F. Costa, and Č. Brukner, Quantum correlations with no causal order, Nature communications 3, 1092 (2012).DOI
  21. [21]C. Branciard, M. Araújo, A. Feix, F. Costa, and Č. Brukner, The simplest causal inequalities and their violation, New Journal of Physics 18, 013008 (2016).DOI
  22. [22]A. Baumeler and S. Wolf, The space of logically consistent classical processes without causal order, New Journal of Physics 18, 013036 (2016).DOI
  23. [23]Ä. Baumeler and S. Wolf, Device-independent test of causal order and relations to fixed-points, New Journal of Physics 18, 035014 (2016).DOI
  24. [24]Ä. Baumeler, F. Costa, T. C. Ralph, S. Wolf, and M. Zych, Reversible time travel with freedom of choice, Classical and Quantum Gravity 36, 224002 (2019).DOI
  25. [25]G. Tobar and F. Costa, Reversible dynamics with closed time-like curves and freedom of choice, Classical and Quantum Gravity 37, 205011 (2020).DOI
  26. [26]H. Dourdent, K. Simonov, A. Leitherer, E.-C. Boghiu, R. Kunjiwa, S. Halder, R. Augustin, and A. Acín, Paradox-free classical non-causality and unambiguous non-locality without entanglement are equivalent (2025), arXiv:2512.23599 [quant-ph].arxiv.org/abs/2512.23599
  27. [27]Ä. Baumeler and E.-E. Tselentis, Equivalence of grandfather and information antinomy under intervention, Electron. Proc. Theor. Comput. Sci. 340, 1–12 (2021).DOI
  28. [28]A. Baumeler, A. Feix, and S. Wolf, Maximal incompatibility of locally classical behavior and global causal order in multipartity scenarios, Physical Review A 90, 042106 M. Araújo and A. Feix, private communication (2014), the process was communicated to Baumeler before it was found by inspecting the extremal points of the non-causal polytope characterized in Baumeler and Wolf [22].
  29. [30]Ä. Baumeler, A. S. Gilani, and J. Rashid, Unlimited non-causal correlations and their relation to non-locality, Quantum 6, 673 (2022).DOI
  30. [31]I. Mejdoub and A. Vanrietvelde, Unilateral determination of causal order in a cyclic process (2025), arXiv:2506.18540 [quant-ph].arxiv.org/abs/2506.18540
  31. [32]H. Dourdent, A. Leitherer, E.-C. Boghiu, K. Simonov, R. Kunjwal, and A. Acín, What makes a causal loop consistent? (2026), arXiv:2609.39735 [quant-ph].arxiv.org/abs/2609.39735
  32. [33]J. Wechs, H. Dourdent, A. A. Abbott, and C. Branciard, Quantum circuits with classical versus quantum control of causal order, PRX Quantum 4, 030901 (2023), [Erratum: PRX Quantum 4, 039001 (2023)], arXiv:2101.08796 [quant-ph].DOI
  33. [34]G. Eftaxias, M. Weilenmann, and R. Colbeck, Multisystem measurements in generalized probabilistic theories and their role in information processing, Physical Review A 108, 062212 (2023).DOI
  34. [35]R. Kunjwal and O. Oreshkov, Nonclassicality in correlations without causal order (2025), arXiv:2307.02565 [quant-ph].arxiv.org/abs/2307.02565
  35. [36]R. J. Greechie, Orthomodular lattices admitting no states, Journal of Combinatorial Theory, Series A 10, 119 (1971).DOI
  36. [37]J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt, Proposed experiment to test local hidden-variable theories, Physical Review Letters 23, 880 (1969).
  37. [38]H. Barnum, C. A. Fuchs, J. M. Renes, and A. Wilce, Influence-free states on compound quantum systems (2005), arXiv:quant-ph/0507108 [quant-ph].DOI
  38. [39]H. Dourdent, Causal and logical loops from quantum theory, Ph.D. thesis, Université Grenoble Alpes (2022).DOI
  39. [40]R. Kunjwal and Ä. Baumeler, Trading causal order for locality, Physical Review Letters 131, 120201 (2023).DOI
  40. [41]T. Fritz, A. B. Sainz, R. Augusiak, J. B. Brask, R. Chaves, A. Levrier, and A. Acín, Local orthogonality as a multipartite principle for quantum correlations, Nature communications 4, 2263 (2013).DOI
  41. [42]S. Pironio, Lifting Bell inequalities, Journal of Mathematical Physics 46, 10.1063/1.1928727 (2005).DOI
  42. [43]R. Augusiak, J. Stasińska, C. Hadley, J. Korbicz, M. Lewenstein, and A. Acín, Bell inequalities with no quantum violation and unextendable product bases, Physical Review Letters 107, 070401 (2011).pubmed.ncbi.nlm.nih.gov/21902377
  43. [44]R. Augusiak, T. Fritz, M. Kotowski, M. Kotowski, M. Pawlowski, M. Lewenstein, and A. Acín, Tight bell inequalities with no quantum violation from qubit unextendible product bases, Physical Review A—Atomic, Molecular, and Optical Physics 85, 042113 (2012).DOI
  44. [45]R. Mothe, A. A. Abbott, and C. Branciard, Correlations and quantum circuits with dynamical causal order (2025), arXiv:2507.07992 [quant-ph].arxiv.org/abs/2507.07992
  45. [46]G. Chiribella, G. M. D’Ariano, and P. Perinotti, Quantum Circuit Architecture, Physical Review Letters 101, 060401 (2008).pubmed.ncbi.nlm.nih.gov/18764438
  46. [47]G. Chiribella, G. M. D’Ariano, and P. Perinotti, Theoretical framework for quantum networks, Physical Review A 80, 022339 (2009).DOI
  47. [48]R. Kunjwal, Beyond the Cabello-Severini-Winter framework: Making sense of contextuality without sharpness of measurements, Quantum 3, 184 (2019).DOI
  48. [49]A. Cabello, Specker’s fundamental principle of quantum mechanics (2012), arXiv:1212.1756 [quant-ph].arxiv.org/abs/1212.1756
  49. [50]E. Specker, Die Logik Nicht Gleichzeitig Entscheidbarer Aussagen, Dialectica 14, 239 (1960), reprinted in E. P. Specker, Selecta (Birkhäuser Verlag, Basel, Switzerland, 1990), pp. 175–182. English translation, by M. P. Seevinck: “The logic of non-simultaneously decidable propositions”, arXiv:1103.4537v3 [physics.hist-ph].
  50. [51]E.-E. Tselentis and Ä. Baumeler, Admissible causal structures and correlations, PRX Quantum 4, 040307 (2023).DOI
  51. [52]C. H. Bennett, D. P. DiVincenzo, C. A. Fuchs, T. Mor, E. Rains, P. W. Shor, J. A. Smolin, and W. K. Wootters, Quantum nonlocality without entanglement, Physical Review A 59, 1070 (1999).DOI
  52. [53]L. Hardy, Quantum theory from five reasonable axioms (2001), arXiv:quant-ph/0101012 [quant-ph].DOI
  53. [54]M. M. Pál, General probabilistic theories: An introduction, Physics Reports 1033, 1 (2023).
  54. [55]A. J. Short and J. Barrett, Strong nonlocality: a trade-off between states and measurements, New Journal of Physics 12, 033034 (2010).DOI
  55. [56]M. F. Pusey, Stabilizer notation for Spekkens’ toy theory, Foundations of Physics 42, 688 (2012).DOI
  56. [57]R. W. Spekkens, Evidence for the epistemic view of quantum states: A toy theory, Physical Review A 75, 032110 (2007).DOI
  57. [58]Y. Chen, P. Wang, and Z. Wang, Causality and Duality in Multipartite Generalized Probabilistic Theories (2024), arXiv:2411.03903 [quant-ph].arxiv.org/abs/2411.03903
  58. [59]K. Sengupta, Achieving Maximal Causal Indefiniteness in a Maximally Nonlocal Theory (2024), arXiv:2411.04201 [quant-ph].arxiv.org/abs/2411.04201
  59. [60]J. Bavaresco, Ämin Baumeler, Y. Guryanova, and C. Budroni, Indefinite causal order in boxworld theories (2024), arXiv:2411.00951 [quant-ph].arxiv.org/abs/2411.00951
  60. [61]Z. Liu and G. Chiribella, Tsirelson bounds for quantum correlations with indefinite causal order, Nature Communications 16, 10.1038/s41467-025-58508-9 (2025).
  61. [62]Z. Liu and O. Oreshkov, Parity erasure: a foundational principle for indefinite causal order (2026), arXiv:2512.08635 [quant-ph].arxiv.org/abs/2512.08635
  62. [63]M. Wilson, Agent policies from higher-order causal functions (2026), arXiv:2512.10937 [cs.AI].arxiv.org/abs/2512.10937
  63. [64]A. Grinbaum, Agency under indefinite causality: operational eternalism in higher-order quantum theory (2026), arXiv:2512.22879 [quant-ph].arxiv.org/abs/2512.22879
  64. [65]Ä. Baumeler, A. S. Gilani, and J. Rashid, Unlimited non-causal correlations and their relation to non-locality, Quantum 6, 673 (2022).DOI
  65. [66]G. Chiribella, G. M. D’Ariano, and P. Perinotti, Theoretical framework for quantum networks, Physical Review A 80, 022339 (2009), arXiv:0904.4483 [quant-ph].DOI
  66. [67]M. Araújo, A. Feix, M. Navascués, and Ć. Brukner, A purification postulate for quantum mechanics with indefinite causal order, Quantum 1, 10 (2017).

Paper details

vlnc:W_3594t6xp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Claude Opus 5.5 (bare)

View statement
Looking for related papers…
Contents