Causal Calculus: The Foundations

Read full text Summarise with VIRAT
Authors: Ojas Gupta
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

While modern machine learning and statistical models excel at pattern recognition, they remain fundamentally trapped on the lowest rung of the ladder of causation: observation. They map correlations but fail to discover the underlying mechanisms that govern reality. Causal Calculus: Volume 1 introduces a unified mathematical framework that bridges Judea Pearl’s do-calculus with differential geometry, topology, and deep learning to create a rigorous "physics of why." This volume transitions causal inference from discrete, static graphs into a continuous, dynamic, and computable space. It outlines the theoretical scaffolding required to build Mechanistic Intelligence—systems capable of intervention, counterfactual reasoning, and causal discovery. Key Contributions & Concepts Explored: The Algebra of Mechanisms: Formalizing how Structural Causal Models (SCMs) can be chained, fused, and represented as Causal Matrices (Jacobians). Differential Causality: Introducing the Causal Gradient, a novel optimization tool that minimizes interventional loss rather than observational error. Causal Geometry & Topology: Defining the Causal Manifold (the continuous space of all possible mechanisms), the Causal Metric Tensor (measuring interventional distance between models), and Causal Holonomy (the mathematical shape of feedback loops and memory). Causal Information Theory: Formulating Causal Entropy and Causal Mutual Information to quantify the exact flow of control and compress bloated models into Minimal Sufficient Mechanisms. The Causal Neural Network (CausalNN): A practical computational architecture that replaces "black-box" dense layers with modular, interpretable causal nodes. The volume concludes with a from-scratch Python/NumPy implementation proving the CausalNN’s ability to learn true physics, perform mechanism surgery, and answer counterfactuals without retraining. This manuscript serves as a foundational text for researchers in artificial intelligence, theoretical computer science, and causal inference who are looking to move beyond curve-fitting and build systems that truly understand cause and effect.

Fields of Study & Subjects

Bayesian Modeling and Causal Inference Logic Reasoning and Knowledge Philosophy of Biology and Scientific Explanation

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN
Document type
Book