Non-Influenceable Causal Order in Quantum Processes

• Non-influenceable causal order is defined as a framework where the distribution of event orders remains independent of local interventions, preventing any party from steering the outcome.
• It establishes a hierarchy that distinguishes between fixed, dynamical, and indefinite causal orders, with quantum switch experiments providing witness-certified evidence of causal non-separability.
• The concept extends beyond quantum circuits to applications in operator-theoretic time-series analysis and non-causal computation, demonstrating its broad relevance across theoretical and experimental frameworks.

Non-influenceable causal order denotes a class of causal structures in which the realized order cannot be steered by parties’ choices or by local interventions. In the causal-correlation literature, a causal correlation $p(\vec a|\vec x)$ has non-influenceable order when there exists a causal decomposition $p(\pi,\vec a|\vec x)$ such that, for every total order $\pi$, the distribution $p(\pi|\vec x)$ is independent of $\vec x$; at the process level, analogous restrictions are imposed on the control weights of higher-order quantum circuits (Mothe et al., 10 Jul 2025). In experimental work on the quantum switch, the same expression is also used operationally for architectures in which no classical side-information—timing, path, or spatial location—can fix or reveal the order on a run-by-run basis, so that witness-certified causal non-separability rules out hidden-variable models with definite order (Goswami et al., 2018).

1. Foundational definitions

The modern formulation of non-influenceable causal order is rooted in the distinction between fixed-order, dynamical, and indefinite causal structures. For $N$ parties $A_1,\ldots,A_N$, a distribution $p(\vec a|\vec x)$ is compatible with a fixed order $\pi=(k_1,\ldots,k_N)$ if every prefix marginal

$p_\pi(a_{k_1},\ldots,a_{k_n}|\vec x)$

is independent of the inputs of the remaining parties. The convex hull of all such fixed-order-compatible sets is denoted $p(\pi,\vec a|\vec x)$0. More general causal correlations allow the order to be established on the fly through a decomposition

$p(\pi,\vec a|\vec x)$1

with linear causal-independence constraints on the relevant marginals. This larger set is the causal polytope $p(\pi,\vec a|\vec x)$2 (Mothe et al., 10 Jul 2025).

In the process-matrix formalism, local operations are represented by Choi operators, and joint probabilities are generated by the generalized Born rule

$p(\pi,\vec a|\vec x)$3

For three events $p(\pi,\vec a|\vec x)$4, a causally separable process has the form

$p(\pi,\vec a|\vec x)$5

Causal non-separability means that no such convex decomposition exists. In the specific quantum-switch experiment of Goswami et al., “non-influenceable causal order” is used in this operational sense: because the two orders occur at the same spatial locations and within the same temporal coherence window, no classical mechanism can set or reveal the order on a run-by-run basis without destroying the coherence that defines the process (Goswami et al., 2018).

The correlation-level definition is stricter than mere dynamicality. A causal correlation has non-influenceable order if there exists a causal decomposition satisfying

$p(\pi,\vec a|\vec x)$6

Thus the order may still be established during the run, but its distribution cannot be steered by earlier parties’ inputs. A weaker notion, coarse-grained non-influenceability, replaces the full order $p(\pi,\vec a|\vec x)$7 by the probability that the next party is $p(\pi,\vec a|\vec x)$8 given the unordered set of past parties $p(\pi,\vec a|\vec x)$9 (Mothe et al., 10 Jul 2025).

2. Correlation-level hierarchy and witnesses

The correlation-level hierarchy separates static order, dynamical but non-influenceable order, and fully general causal order. The relevant inclusions are

$\pi$0

These inclusions collapse in low-party scenarios: for $\pi$1,

$\pi$2

and for $\pi$3,

$\pi$4

Only for $\pi$5 do the inclusions become strict, which is the precise sense in which dynamicality and influenceability separate as distinct resources (Mothe et al., 10 Jul 2025).

This separation is witnessed by linear inequalities in the lazy multipartite scenarios. In the four-party lazy scenario, the quantity

$\pi$6

obeys the strict bounds

$\pi$7

In the tripartite lazy scenario, the simpler facet

$\pi$8

distinguishes static from dynamical order, since causal correlations can reach $\pi$9 when the order between the last two parties depends on the first party’s input (Mothe et al., 10 Jul 2025).

The sets $p(\pi|\vec x)$0 and $p(\pi|\vec x)$1 are convex polytopes defined by linear constraints. Membership is therefore reducible to linear-program feasibility: one searches for a causal decomposition $p(\pi|\vec x)$2 obeying causal constraints, normalization, nonnegativity, and either $p(\pi|\vec x)$3 or its coarse-grained analogue. This makes non-influenceability a structurally sharp property rather than an informal prohibition on “controlling the order” (Mothe et al., 10 Jul 2025).

3. Quantum processes, circuit classes, and realizability

At the process level, non-influenceability is formalized through subclasses of higher-order circuits with restricted control registers. Quantum circuits with classical control of order (QC-CC) generate causal correlations, but the non-influenceable subclass QC-NICC requires that all prefix blocks $p(\pi|\vec x)$4 be valid process matrices up to normalization. Quantum circuits with quantum control of order (QC-QC) similarly admit a non-influenceable subclass QC-NIQC, defined by requiring that all blocks $p(\pi|\vec x)$5 be valid process matrices up to normalization. These classes satisfy, for $p(\pi|\vec x)$6,

$p(\pi|\vec x)$7

and

$p(\pi|\vec x)$8

with the corresponding correlation images

$p(\pi|\vec x)$9

For $\vec x$0, the non-influenceable subclasses collapse back to static-order ones (Mothe et al., 10 Jul 2025).

A closely related top-down characterization is provided by the causal-box analysis of process boxes. In a classical acyclic spacetime, if one imposes the fine-grained closed-laboratory constraints “Acting Once” and “Local Order,” then any such protocol is behaviorally equivalent to a quantum circuit with quantum control of causal order, and conversely. The resulting equivalence

$\vec x$1

shows that QC-QCs are precisely the physically realizable higher-order quantum processes, including indefinite-order ones, within classical spacetime under closed-lab assumptions (Salzger et al., 8 May 2026).

This perspective places a sharp limit on stronger notions of non-influenceability. Costa’s no-go theorem shows that a pure, control-free superposition

$\vec x$2

of differently ordered unitary Markovian pure processes with equal local dimensions is not a valid deterministic process vector. Coherent control therefore implies that some ancillary system carries which-order information; if that control is traced out, the order reduces to a probabilistic mixture rather than a pure superposition (Costa, 2020).

A complementary constraint appears in constructive simulations of acausal processes. In the non-Markovian circuit-with-measurement construction of Feix et al., a conditioned process is proper if and only if the success probability $\vec x$3 is independent of the parties’ instruments. This provides a local non-influenceability criterion for postselected realizations: if the postselection rate depended on the instruments, the reconstructed process would fail the local-causality conditions required of a valid process matrix (Milz et al., 2017).

4. Quantum switch as the canonical operational example

The most developed operational realization is the quantum switch. In the photonic implementation of Goswami et al., the target is the photon’s transverse spatial-mode qubit with basis $\vec x$4, and the control is the polarization qubit $\vec x$5. The switch implements

$\vec x$6

For control state $\vec x$7, the output is

$\vec x$8

which is a coherent superposition of the two possible orders (Goswami et al., 2018).

The experimental claim of non-influenceability rests on architectural indistinguishability. Both orders occur at the same two fixed spatial locations regardless of the control state, and the photons’ coherence length is approximately $\vec x$9 whereas the interferometer path separation is approximately $N$0. Consequently, the operations cannot be distinguished, even in principle, by spatial position or timing. This removes the order-revealing side channels that affected earlier black-box realizations (Goswami et al., 2018).

Certification is performed by a causal witness. For causally separable processes,

$N$1

while the witness used in the experiment admits the decomposition

$N$2

with experimentally accessible expectation value

$N$3

The measured value was

$N$4

which is $N$5 standard deviations below the definite-order bound $N$6; with a conservative corrected bound $N$7 accounting for rotational misalignment, the violation remains $N$8 (Goswami et al., 2018).

This experiment followed the first decisive witness-based demonstration of indefinite causal order, where Procopio et al. measured a causal witness corresponding to $N$9, certifying causal non-separability by almost seven standard deviations (Rubino et al., 2016). The later photonic architecture strengthened the operational reading of non-influenceability: not only was the process causally non-separable, but the order also could not be classically detected or fixed through path or timing information (Goswami et al., 2018).

5. Device-independent tests and recurrent misconceptions

A central controversy concerns whether violations of fixed-order inequalities certify indefinite causal order. The answer is negative in general. In the $A_1,\ldots,A_N$0-cycle game, adaptive control by a global-past party can select the active link, and both classical and quantum switches reach $A_1,\ldots,A_N$1; the quantum switch violates the fixed-order inequality in this regime without exploiting its indefinite nature. When the past and future parties are required to act non-adaptively, however, the dynamical aspect is removed, and for $A_1,\ldots,A_N$2 the fixed-order $A_1,\ldots,A_N$3-cycle inequality becomes a causal inequality. In that non-influenceable regime, switch-like processes cannot violate it (Baumann et al., 2024).

This distinction aligns with the broader separation between causal non-separability and noncausality. QC-QCs, including the quantum switch, are causally non-separable but generate only causal correlations. In the SHIFT–Lugano constructions, successful SHIFT discrimination implemented through a quantum switch of classical communications witnesses causal non-separability of the induced distributed measurement, not device-independent noncausality; the local laboratories cannot influence the control, and the resulting correlations remain inside the causal polytope (Steffinlongo et al., 21 Feb 2025).

Fully device-independent certification nonetheless became possible in extended switch scenarios. The van der Lugt–Barrett–Chiribella inequality involves two operations inside the switch, a future party, and a spacelike separated party. Its definite-order bound is

$A_1,\ldots,A_N$4

An experimental implementation measured

$A_1,\ldots,A_N$5

which is $A_1,\ldots,A_N$6 standard deviations above the classical bound $A_1,\ldots,A_N$7, although standard and ICO-specific loopholes remained open (Richter et al., 20 Jun 2025).

The strongest current device-independent formulation is possibilistic and maximal. Using GHZ- and Mermin-type arguments, Gogioso and Pinzani showed that, under the impossibility of superluminal influences, the observed correlations of suitably arranged quantum switches are incompatible even with the existence of a causal order on only a small fraction of runs. In the ideal probabilistic limit, the associated causal Mermin inequality is saturated algebraically by the switch, implying that no positive fraction of runs can carry a definite causal order consistent with the no-superluminal-influence constraints (Lugt et al., 2023).

6. Extensions and alternative uses of the concept

Outside the standard process-matrix setting, the phrase has acquired more specialized meanings. In operator-theoretic multivariate time-series analysis, non-influenceable causal order from source component $A_1,\ldots,A_N$8 to target set $A_1,\ldots,A_N$9 is defined as exact invariance of orthogonally invariant spectral functionals under admissible order-preserving deformations of the source. In the frequency domain the null is

$p(\vec a|\vec x)$0

and in the lag domain it is equivalent to constancy of

$p(\vec a|\vec x)$1

over $p(\vec a|\vec x)$2. The test statistic

$p(\vec a|\vec x)$3

is calibrated by shift-based randomization exploiting order-induced group invariance. Under linear Gaussian VAR assumptions, this criterion coincides with Granger noncausality (Dominguez, 3 Jan 2026).

In higher-order boxworld, the closest analogue is the principle “no signaling without system exchange.” There, non-influenceability means that nonsignaling local operations cannot induce signaling correlations. Together with nonsignaling preservation, this produces a nontrivial higher-order theory that excludes perfect two-way signaling yet attains causal-inequality violations larger than those known in higher-order quantum theory, including $p(\vec a|\vec x)$4 for the OCB inequality (Bavaresco et al., 2024).

In non-causal computation, the term appears in an emergent fixed-point sense. Global causal order is absent, but logical consistency requires a unique fixed point $p(\vec a|\vec x)$5. Once such a circuit is specified, the induced dependency structure is non-influenceable by further local interventions that preserve consistency: any attempted reordering or alteration that would change the fixed point produces overdetermination or underdetermination and is excluded (Baumeler et al., 2016).

Across these settings, non-influenceable causal order does not denote a single universal formal object. Rather, it marks a family of constraints forbidding parties from steering, revealing, or operationally reprogramming causal structure by local means. In the multipartite causal-correlation hierarchy it identifies an intermediate resource between static and fully dynamical causal order; in switch experiments it becomes an operational synonym for witness-certified causal non-separability under stringent indistinguishability conditions; in device-independent analyses it is tied to no-superluminal-influence assumptions; and in broader operator-theoretic or generalized-probabilistic settings it is recast as invariance under admissible deformations or as the impossibility of generating signaling without genuine system exchange.