A Law of Emergence: Maximum Causal Power at the Mesoscale
Liang Chen · 2025
Read A Law of Emergence: Maximum Causal Power at the Mesoscale by Liang Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-12016.txt.
Introduction
Emergence, where large-scale behavior grows out of small local interactions, appears in magnets, flocks, and markets, but no law has said where it peaks. Liang Chen proposes one. He defines a system's causal power at each spatial scale as the information an intervention at that scale carries about the outcome, then proves a Middle-Scale Peak Theorem.
For a broad class of locally interacting systems, causal power is not steady across scales: it rises to a strict peak at a middle scale. The peak comes from a trade-off between noise averaging at small scales and limited reach at large ones. Tests on a 2D Ising model near criticality and on an agent model of collective behavior confirmed the predicted peak. Chen offers the law as a falsifiable foundation.
Catalog record
| CREATOR | Liang Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-12016
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Law of Emergence: Maximum Causal Power at the Mesoscale Liang Chen Abstract Complex systems universally exhibit emergence, where macroscopic dynamics arise from local interactions, but a predictive law governing this process has been absent. We establish and verify such a law. We define a system’s causal power at a spatial scale, ℓ \ell , as its Effective Information (EI ℓ ) , measured by the mutual information between a targeted, maximum-entropy intervention and its outcome. From this, we derive and prove a Middle-Scale Peak Theorem : for a broad class of systems with local interactions, EI ℓ is not monotonic but exhibits a strict maximum at a mesoscopic scale ℓ ∗ \ell^{*} . This peak is a necessary consequence of a fundamental trade-off between noise-averaging at small scales and locality-limited response at large scales. We provide quantitative, reproducible evidence for this law in two distinct domains: a 2D Ising model near criticality and a model of agent-based collective behavior. In both systems, the predicted unimodal peak is decisively confirmed by statistical model selection. Our work establishes a falsifiable, first-principles law that identifies the natural scale of emergence, providing a quantitative foundation for the discovery of effective theories. Keywords: Causal Emergence, Complex Systems, Information Theory, Statistical Physics, Effective Theories, Multi-scale Modeling. 1 Introduction 1.1 The Enduring Quest for the Right Scale A foundational challenge in science is identifying the appropriate level of description for a given phenomenon. The history of science can be viewed as a dynamic tension between reductionism—the belief that systems are best understood by dissecting them into their constituent parts—and holism, which posits that the whole is more than the sum of its parts. This tension was famously articulated by P.W. Anderson in his 1972 essay ”More is Different” [ 10 ] , which argued that the hierarchical structure of science, from particle physics to social science, is characterized by the emergence of new, irreducible laws at each level of complexity. This principle underpins our understanding of phenomena from phase transitions in condensed matter physics [ 20 ] to the collective intelligence of biological organisms [ 18 ] . Yet, a fundamental question has remained unresolved: At what scale, if any, is a system’s behavior most causally potent? The absence of a formal, predictive law to answer this has left the study of emergence a largely qualitative, and often controversial, field. 1.2 From Observational to Interventional Frameworks Previous frameworks have laid critical groundwork. The renormalization group (RG) provides a mathematical formalism for understanding how physical laws change with scale, explaining the emergence of universal behavior by systematically integrating out microscopic details [ 15 , 20 ] . Computational mechanics identifies optimally predictive representations by constructing minimal ”causal states” from observational time-series data [ 17 ] . More recently, theories of causal emergence have used information theory to show that macro-variables can sometimes be more informative or causally effective than their micro-foundations [ 12 ] . More recently, theories of causal emergence have used information theory to show that macro-variables can sometimes be more informative or causally effective than their micro-foundations [ 12 ] . This approach has been extended through information decomposition methods [ 1 ] and shown to be robust across multiple causation measures [ 4 ] . Recent surveys highlight the rapid development of this field [ 2 , 3 ] . However, a complete theory of effective description must be interventional [ 16 ] . The ultimate test of a scale’s utility is not merely passive prediction, but active control. To claim that a macroscopic variable, like pressure, is a ”real” and useful feature of a gas, one must be able to manipulate it (e.g., with a piston) and observe a reliable outcome. This paper places the concept of intervention at the heart of the problem. 1.3 A Variational Principle for Emergence Conceptually, we posit that emergence is a process of optimization. We propose that natural systems, through evolutionary or dynamical processes, tend to find descriptive levels that maximize their causal power while minimizing complexity. This can be formalized through a variational principle, where a system’s macro-trajectory m(t) extremizes an action \mathcal{S}[m] : \mathcal{S}[m]=\int\left(\underbrace{\mathcal{I}_{\text{cause}}(m)}_{\text{causal efficacy}}-\lambda\,\underbrace{\mathcal{C}(m)}_{\text{descriptive complexity}}\right)dt. (1) While a full exploration of this principle is beyond our current scope, it provides a powerful conceptual lens. Our work provides a direct, testable consequence of this picture by focusing on the causal efficacy term, \mathcal{I}_{\text{cause}} , demonstrating that it is maximized at a specific, non-trivial scale dictated by the system’s physical constraints. 1.4 Our Contribution This paper addresses the challenge of finding the optimal scale by introducing a formal, interventional framework. Our contribution is threefold: 1. We provide a formal, operational, and scalable definition of scale-dependent causal power , the Effective Information EI ℓ . 2. We derive and prove the Middle-Scale Peak Theorem , which predicts a unimodal dependence of EI ℓ on the scale \ell from first principles. 3. We provide reproducible, cross-domain computational evidence for the theorem, using a unified analysis pipeline on a physical (Ising) and a biological (agent-based) model system. Our work establishes a law-like, falsifiable criterion for identifying the scale of maximal causal power, transforming emergence from a qualitative concept into a quantitative science. 2 An Interventional Framework for Causal Power 2.1 Systems, Scales, and Coarse-Graining Definition 1 (System and Scales) . Consider a system whose microstates x\in\mathcal{X} evolve according to a discrete-time Markov process with transition kernel K(x^{\prime}|x) . A hierarchy of scales is defined by a set of coarse-graining maps T_{\ell}:\mathcal{X}\to\mathcal{M}_{\ell} , where M_{t}\equiv T_{\ell}(X_{t}) is the macrostate at scale \ell . In this work, for a system on a lattice, \ell corresponds to the linear size of a block used for averaging. 2.2 The Causal Probe: Maximum-Entropy Interventions To compare causal efficacy across scales, we need a standardized probe. The choice of intervention is critical; an ill-chosen probe might reveal more about the probe itself than the system. We therefore adopt the principle of maximum entropy. Definition 2 (Maximum-Entropy Intervention) . A Maximum-Entropy (MaxEnt) intervention, denoted \mathrm{do}_{\ell}[X_{t}] , sets the microstate distribution p(x) to be the one that maximizes Shannon entropy H[p] subject to a macroscopic constraint on the distribution of M_{t}=T_{\ell}(X_{t}) . This choice is crucial as it represents a maximally unbiased probe. It tests the system’s intrinsic causal structure by setting a macro-variable to a desired state while imposing no additional, arbitrary correlations at the micro-level. It is the informational equivalent of ”doing the least” to achieve a macroscopic goal. 2.3 Quantifying Causal Power: Effective Information at Scale With a standardized probe defined, we can now quantify the causal power of a given scale. We define this as the amount of information transmitted from our intervention at time t to the system’s state at time t+\Delta t . Definition 3 (Effective Information at Scale) . The causal power at scale \ell , or Effective Information at Scale (EI ℓ ) , is the mutual information between the system’s macrostate at time t and t+\Delta t under a uniform MaxEnt intervention on M_{t} : \mathrm{EI}_{\ell}\equiv I_{\mathrm{do}_{\ell}}(M_{t};M_{t+\Delta t}). (2) EI ℓ quantifies the degree of reliable control one has over the system’s future by manipulating its present at scale \ell . For spatially extended systems, we report