Causal Inequality in Quantum Systems

Updated 23 June 2026

Causal inequality is a principle defining operational bounds on correlations produced under a fixed, definite causal order in localized quantum operations.
The process-matrix framework underpins its derivation, using linear constraints to delineate classical, quantum, and post-quantum causal structures.
Experimental violations, demonstrated via photonic quantum switches, open avenues for leveraging indefinite causal order in quantum information processing.

Causal inequality is a fundamental concept at the intersection of quantum foundations and causal inference, expressing operational constraints on correlations that arise under the assumption of definite causal order. These inequalities embody the device-independent limits on the observable joint distributions generated by parties conducting localized operations, typically in the presence of possible latent variables but absent any superpositions or indefiniteness of causal structure. The violation of a causal inequality thus serves as a signature of scenarios where fixed background causal ordering—classical or quantum—cannot explain the observed data. Causal inequalities have played a central role in delineating the boundary between classical, quantum, and “post-quantum” causal structures, with profound implications for quantum information, experimental physics, and the logical foundations of causality.

1. Causal Inequality: Definition and Origin

The archetypal causal inequality was introduced in the context of the process-matrix framework developed by Oreshkov, Costa, and Brukner (Oreshkov et al., 2011). Consider two separated parties, Alice (A) and Bob (B), each within a closed local laboratory. Each receives an incoming quantum system, performs a quantum instrument (CP, trace non-increasing map), records classical outcomes, and emits an outgoing system. In each run, Alice generates a random bit $a$ , and Bob independently generates bits $b$ and $b'$ . The experimental task is:

If $b'=0$ : Alice attempts to guess Bob's bit $b$ , outputting guess $x$ .
If $b'=1$ : Bob attempts to guess Alice's bit $a$ , outputting guess $y$ .

With random, uncorrelated choices and the additional assumptions of locally valid quantum mechanics, free choice, and laboratory isolation, all systems respecting a definite (possibly unknown) causal order — i.e., for which events can be embedded in a background partial order — must satisfy

$p_{\rm succ} = \tfrac{1}{2}\left[P(x=b \mid b'=0) + P(y=a \mid b'=1)\right] \leq \tfrac{3}{4}$

This bound is the original "causal inequality."

2. Derivation and Causal Polytope

The derivation proceeds by examining the three mutually exclusive causal relations between the entry events $b$ 0 (system enters Alice's lab) and $b$ 1 (Bob's): (a) $b$ 2 (Alice causally before Bob), (b) $b$ 3, or (c) the events are causally unrelated (e.g., space-like separated). By the closed-lab assumption, the outputs $b$ 4 and $b$ 5 can only depend on inputs in their causal past. For each possible order, only one direction of perfect signalling is allowed, with the other direction being limited to random guessing. Averaging over possible (even probabilistic) mixtures, the convex structure yields the polytope of causal correlations, and the facet-defining linear constraints are the causal inequalities: e.g., the $b$ 6 bound in the above scenario (Oreshkov et al., 2011, Branciard et al., 2015).

Table: Key Specification in the Oreshkov–Costa–Brukner Causal Game

Party operation	Guessing Task	Maximum Prob. (per definite order)
$b$ 7	$b$ 8=1: $b$ 9	1 (Bob knows $b'$ 0), $b'$ 1 at random
$b'$ 2	$b'$ 3=0: $b'$ 4	1 (Alice knows $b'$ 5), $b'$ 6 at random
Mix	-	Average $b'$ 7

The causal polytope—the convex hull of all correlations compatible with definite orderings—fully captures all permissible joint distributions for this scenario (Branciard et al., 2015).

3. Quantum Violations and the Process-Matrix Framework

The process-matrix framework relaxes the assumption of a predefined global time or causal structure, postulating instead only that quantum mechanics applies locally inside each lab. Here, correlations are specified by a positive semidefinite “process matrix” $b'$ 8 acting on the tensor product of all lab input and output spaces. Joint probabilities are given by

$b'$ 9

for Choi representations $b'=0$ 0 of valid CP maps. For specific non-separable $b'=0$ 1, one can construct quantum protocols that achieve

$b'=0$ 2

thus violating the classical bound of $b'=0$ 3 (Oreshkov et al., 2011, Brukner, 2014).

In these protocols, the causal structure underlying events is itself in a quantum superposition—not merely “randomized” but fundamentally indefinite. This has operational consequences, conferring advantages in channel discrimination, communication, and certain computation tasks (Brukner, 2014, Guo et al., 25 Jun 2025).

4. Device-Independent Experimental Violations

Recent experimental realizations have demonstrated statistically significant violations of causal inequalities in controlled settings. In such experiments, a photonic quantum switch is typically implemented, wherein the order of two operations depends coherently on the state of a control qubit, itself entangled or spacelike separated from other degrees of freedom (Richter et al., 20 Jun 2025, Guo et al., 25 Jun 2025, Qu et al., 6 Aug 2025). The paradigmatic “VBC inequality” (van der Lugt, Barrett, Chiribella) involves four parties and a composite scenario:

$b'=0$ 4

Quantum theory predicts $b'=0$ 5, and experiments have been reported measuring $b'=0$ 6 (24 standard deviations above the classical bound) (Richter et al., 20 Jun 2025, Qu et al., 6 Aug 2025). This constitutes, in analogy to Bell tests, a device-independent witness to indefinite causal order.

The structure of causal inequalities generalizes to settings with more parties, arbitrary input/output cardinality, and biased inputs (Branciard et al., 2015, Bhattacharya et al., 2015). The set of classical models forms a polytope whose facets define all possible causal inequalities for any given scenario. In the bipartite binary-input/output case, there exist two nonequivalent families: the GYNI-type ( $b'=0$ 7) and “lazy GYNI” ( $b'=0$ 8) (Branciard et al., 2015).

Furthermore, adopting the device-independent approach, one can define the notion of “causal nonseparability” for process matrices, and constrain quantum violations by Tsirelson-like bounds; e.g., maximal quantum value of $b'=0$ 9 under valid (traceless, dichotomic) observables is $b$ 0 (Brukner, 2014, Bhattacharya et al., 2015).

In the causal-inference literature, related types of “causal inequalities” emerge as testable implications of graph-structured models with latent variables, especially in instrumental-variable settings. These take the form of polynomial or linear constraints on observable distributions, narrowing the set of compatible causal models (Pearl, 2013, Kang et al., 2012, Bhadane et al., 14 Feb 2025, Finkelstein et al., 2020). The instrumentality inequalities introduced by Pearl, and their extensions, serve as practical diagnostic tools in nonparametric identification, fairness testing, and model falsification.

6. Physical and Foundational Significance

Causal inequality violations play an analogous role to Bell-inequality violations in nonlocality. However, whereas Bell nonlocality challenges classical realism and the notion of predetermined outcomes, causal-inequality violation challenges the notion of an absolute, global order of cause and effect—suggesting that quantum theory admits “superpositions of causal orders.” In the classical (pointer basis) or decohered limit, definite causal relations and the corresponding polytope constraints are restored (Oreshkov et al., 2011). This supports a view in which spacetime and macroscopic causality emerge from fundamentally indefinite, quantum-gravitational processes.

Operationally, the ability to violate causal inequalities is now recognized as a genuine quantum resource, opening the possibility of protocols and devices exploiting indefinite order for quantum advantage in information processing (Guo et al., 25 Jun 2025).

7. Controversies, Limitations, and Open Directions

While quantum theory, as formalized by the process-matrix approach, can violate causal inequalities, not every generalized causal scenario admits such violations. In particular, it has been shown that any experiment implementable within standard circuit-based quantum theory, even if coherently controlling the order of operations, still respects all causal inequalities—a property not shared with the full space of process matrices (Purves et al., 2021). This distinction pinpoints the process-matrix framework as strictly more powerful than quantum circuits with coherent control, and motivates further research into finding physical realizations beyond the quantum switch.

Moreover, in quantum field-theoretic settings, causal inequality violation may occur even in Minkowski spacetime due to the operational nonlocality of field modes, raising subtle questions about the assumptions underlying “closed laboratories” (Ho et al., 2018).

A systematic classification of causal scenarios according to their algebraic (CI only) or non-algebraic (inequality-constrained) correlational structure is advancing, but comprehensive necessary and sufficient conditions for the presence of causal inequalities beyond four observed variables remain unresolved (Khanna et al., 2023). Experimental loophole closure in device-independent indefinite-order certification remains an open technical challenge (Qu et al., 6 Aug 2025).

In summary, causal inequalities demarcate the boundary between correlations explainable by definite causal order (classical or quantum) and those that are not. Their violation—both theoretically and experimentally—heralds new classes of quantum resources and poses challenging questions for our deepest understanding of space, time, and information processing (Oreshkov et al., 2011, Brukner, 2014, Guo et al., 25 Jun 2025, Richter et al., 20 Jun 2025, Qu et al., 6 Aug 2025).