Mathematical Foundations for Unified Causal Inference

This direction is split into other directions: intervention-free actual cause definition, causal-logical equivalence, parameterized causal unification. Its last version stays readable and citable here.

Version 1 of 153 papers65 conceptsfrozen 2026-10-04

Causal inference has historically been a fragmented field. Depending on whether you are an econometrician, a computer scientist, or a philosopher, you might represent "cause" using potential outcomes, directed acyclic graphs (DAGs), or modal logic. While these frameworks often reach the same conclusions, their mathematical languages are distinct, making it difficult to transfer insights between them or to handle complex scenarios like cyclic dependencies and infinite variables.

A growing body of research seeks to provide a "Rosetta Stone" for causality—a unified mathematical foundation that treats these disparate models as special cases of a single, more expressive meta-language.

Logic as a Superset for Equations

The most common way to represent causality is through Structural Causal Models (SCMs), which use algebraic equations to describe how variables determine one another. However, Alexander Bochman and Vladimir Lifschitz demonstrate in Pearl's Causality in a Logical Setting that these equations can be translated into nonmonotonic logic. In this view, the "solutions" to causal equations are simply the logical models of a theory.

This logical framing is more than just a translation; it expands what we can model. For instance, Chitta Baral and Nam Tran show in Encoding Probabilistic Causal Model in Probabilistic Action Language that formal action languages serve as a superset for Pearl’s models. While SCMs struggle with "ramifications" (indirect effects) and "qualifications" (conditions that prevent an action), logical frameworks handle them naturally by treating them as manifestations of the same underlying causal structure, as argued by Norman McCain and Hudson Turner in A Causal Theory of Ramifications and Qualifications.

Category Theory and Universal Causality

If logic provides the grammar for causality, category theory provides the geometry. Sridhar Mahadevan has proposed a framework called Universal Causality (UC). In On The Universality of Diagrams for Causal Inference and The Causal Reproducing Property, Mahadevan uses the Yoneda Lemma to prove that all causal influences can be represented as "diagrams" in a category.

This approach works because it abstracts away the implementation of a causal effect (whether it's a linear regression or a complex simulation) and focuses on the morphisms (the relationships) between objects. By moving to Topos Causal Models (TCMs), as detailed in Topos Causal Models, researchers can use the internal logic of a topos to reason about counterfactuals. This reveals that causal logic is fundamentally intuitionistic rather than Boolean—meaning that in complex systems, the "truth" of a causal claim depends on the available evidence and the structure of the system, rather than a simple binary state.

From Variables to Measure-Theoretic Events

Traditional causal models rely on discrete variables (e.g., $X$ causes $Y$). But real-world systems, like climate models or high-frequency trading, often involve continuous time and infinite variables. Joseph Halpern and Spencer Peters address this in A Unifying Framework for Causal Modeling With Infinitely Many Variables by introducing Generalized Structural-Equations Models (GSEMs).

GSEMs shift the focus from specific equations to an "interface" of interventions and outcomes. This allows researchers to apply causal definitions to dynamical systems (like ODEs) while maintaining mathematical equivalence to standard models in finite settings. Similarly, Junhyung Park and colleagues propose a Measure-Theoretic Axiomatisation of Causality. By building causality on measure theory, they treat "causes" as sets of events in a sigma-algebra. This unification shows that traditional variable-based causality is actually a restricted application of a more universal measure-theoretic approach, as explored in A fine-grained look at causal effects in causal spaces.

Reconciling Actual Causation

A major point of friction in the field is Actual Causation: the study of whether a specific event caused a specific outcome (e.g., "Did the driver's text cause this accident?"). Historically, this was split between "regularity" theories (looking for sufficient sets of conditions) and "counterfactual" theories (asking "what if?").

Sander Beckers and Joost Vennekens provide a bridge in Towards a General Framework for Actual Causation Using CP-logic. By parameterizing which causal laws are "intrinsic" to a situation, they show that competing philosophical definitions are actually just different configurations of the same logical structure. This culminated in the Counterfactual NESS (CNESS) definition in The Counterfactual NESS Definition of Causation, which formalizes the legal "Necessary Element of a Sufficient Set" test within the mathematical framework of SCMs.

Why Unification Works

The power of these unified frameworks lies in their ability to treat observational data as a special case of experimental data. As Sanghack Lee and Elias Bareinboim demonstrate in General Identifiability with Arbitrary Surrogate Experiments, once you have a unified representation, you can calculate causal effects from any combination of data sources using a single algorithm.

By stripping away the idiosyncratic notation of different subfields, this research direction reveals that causality is not a collection of disparate tools, but a singular mathematical property of how information and influence flow through a system.

Papers in this version

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_mathematical_foundations_for_unified_cau_f1ee295a,
  title = {Mathematical Foundations for Unified Causal Inference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/mathematical-foundations-for-unified-causal-inference/rdr_f1ee295a9c3d42219915ed5027e68e15},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_mathematical_foundations_for_unified_cau_f1ee295a_v1,
  title = {Mathematical Foundations for Unified Causal Inference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/mathematical-foundations-for-unified-causal-inference/rdr_f1ee295a9c3d42219915ed5027e68e15/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}