Characterizing Signalling: Connections between Causal Inference and Space-time Geometry

Technical Summary: Characterizing Signalling: Connections between Causal Inference and Space-time Geometry

Problem Statement
Causality is understood through two distinct lenses: information-theoretic causality, defined by the flow of information and inferable through agent interventions, and relativistic causality, defined by the geometric structure of space-time (specifically light cones). While these notions must be compatible in physical theories, their interplay is complex, particularly when "fine-tuning" is present. Fine-tuning occurs when causal mechanisms are adjusted such that observable correlations or signalling vanish despite underlying causal connections (unfaithful causal models).

A critical open question is how the relativistic principle of No Superluminal Signalling (NSS) constrains information-theoretic causal structures. Specifically, does the absence of superluminal signalling in a space-time with an acyclic causal structure guarantee the absence of causal loops in the information-theoretic model? Previous work [11] demonstrated that in 1+1-dimensional Minkowski space-time, cyclic causal models can be embedded compatibly with NSS, implying that NSS does not preclude causal loops in lower dimensions. However, it remains unclear whether this holds for higher dimensions ($d > 1$) and what geometric or informational properties distinguish these cases. Furthermore, standard causal inference techniques often fail in the presence of fine-tuning, complicating the operational identification of causal structures.

Methodology
The authors build upon the "affects framework" introduced in [10], a theory-independent causal modelling approach that formalizes signalling through "affects relations." An affects relation $X \models Y \mid \text{do}(Z)$ signifies that an intervention on $X$ allows an agent to signal to an observer of $Y$, given knowledge of interventions on $Z$.

The paper employs the following methodological steps:

1. Refinement of Affects Relations: The authors extend the concept of "reducibility" (identifying redundant variables in an affects relation) to all arguments of the relation (interventional $X, Z$ and observational $Y$). They introduce "clustering," a property where an affects relation holds for a set of variables but not for any of its strict subsets.
2. Order-Theoretic Analysis of Space-time: Space-time is modeled as a partially ordered set (poset). The authors introduce a new order-theoretic property called "conicality," which relates the joint future of a set of points to the unique set of "spanning" points that generate that future.
3. Compatibility Analysis: The authors analyze the conditions under which a set of affects relations can be embedded into a space-time without violating NSS. This involves checking if the joint future of the observational variables lies within the intersection of the futures of the interventional variables.
4. Correspondence Proofs: The paper proves theorems establishing a correspondence between the ordering constraints derived from causal inference (information-theoretic) and those derived from NSS (geometric) under specific conditions.

Key Contributions and Results

1. Characterization of Signalling and Fine-tuning:

   • Reducibility and Clustering: The authors define reducibility for all arguments of an affects relation. They prove that "clustering" (the absence of affects relations for subsets) is a signature of fine-tuning (Theorem 4.12).
   • Causal Inference in Unfaithful Models: A significant result is that in certain fine-tuned (unfaithful) models, the absence of signalling between specific nodes can be used to infer causal connections. Specifically, if an affects relation is irreducible in the first and third arguments, every node in the interventional sets $X$ and $Z$ is a cause of some node in $Y$ (Corollary 4.15). This allows for causal inference even when standard faithfulness assumptions are violated.
   • Classification of Fine-tuning: The work distinguishes between "observational fine-tuning" (clustering in the second argument, e.g., jamming scenarios) and "interventional fine-tuning" (clustering in the first or third arguments).

2. Order-Theoretic Property: Conicality:

   • The authors define a poset as "conical" if the joint future of any finite set of points uniquely determines the set of spanning points contributing to that future.
   • Lemma 5.4: They prove that Minkowski space-times with $d > 1$ spatial dimensions are conical, whereas 1+1-dimensional Minkowski space-time is not. In $d=1$, distinct pairs of points can share the same joint future, violating conicality.

3. Correspondence between Causal Inference and Space-time Geometry:

   • The main result (Theorems 6.3 and 6.4) establishes that the ordering constraints imposed by NSS and those imposed by causal inference align (i.e., the interventional arguments are strictly "before" the observational arguments in both the causal graph and the space-time light cone structure) under two specific scenarios:
     1. Faithful Models: Any causal model that is not fine-tuned (no clustering) embedded in an arbitrary acyclic space-time.
     2. Conical Space-times: Any causal model (even fine-tuned ones) embedded in a conical space-time.
   • In these scenarios, an irreducible affects relation $X \models Y \mid \text{do}(Z)$ implies that the joint future of $Y$ is contained within the intersection of the futures of $X$ and $Z$ ($\bar{F}_s(Y) \subseteq \bar{F}_s(X) \cap \bar{F}_s(Z)$).
   • Counter-example: In non-conical space-times (like 1+1 Minkowski) with fine-tuned models, this alignment fails; NSS does not guarantee the causal order derived from inference (Example 11).

Significance and Claims
The paper claims to provide a rigorous formalization of the relationship between informational and geometric causality. Its primary significance lies in:

• Resolving the Dimensional Discrepancy: It offers a formal explanation for why causal loops compatible with NSS exist in 1+1 dimensions but are conjectured to be impossible in higher dimensions. The property of "conicality" distinguishes these geometries, and the correspondence theorems suggest that in conical space-times, NSS is sufficient to rule out operationally detectable causal loops for faithful models.
• Tools for Causal Inference: By characterizing reducibility and clustering, the work provides tools to perform causal inference in unfaithful models, a regime where standard algorithms typically fail. It demonstrates that the absence of signalling can be a diagnostic tool for causal structure.
• Unifying Framework: The results bridge the gap between the "top-down" affects framework and the geometric constraints of relativity, showing that the principles of NSS and no causal loops align under specific geometric (conicality) or informational (faithfulness) conditions.

The authors conclude that their findings support the conjecture that in conical space-times (such as our physical 3+1 universe), the principle of no superluminal signalling is sufficient to rule out operationally detectable causal loops, provided the causal models are faithful. For fine-tuned models, the paper suggests that the possibility of embedding such loops in higher dimensions may require a specific "fine-tuning" of the space-time embedding itself, analogous to fine-tuning in causal mechanisms.