The Inaccessible Game

Neil D. Lawrence · 2025

Read The Inaccessible Game by Neil D. Lawrence (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-2511-06795.txt.

Introduction

Neil D Lawrence introduces the inaccessible game, an information-theoretic model of a dynamical system built from four axioms. The first three axioms are known results that define information loss within the system. The fourth is new: an information isolation axiom stating that the system is cut off from observation, making it observer-independent.

Under that isolation axiom, the total marginal entropy of the system stays constant rather than growing or shrinking. Lawrence then studies maximum entropy production inside the game and shows the dynamics take a structure resembling the GENERIC system of nonequilibrium physics. That structure combines a reversible part and an irreversible part, mirroring how physicists describe systems that mix mechanics with dissipation.

Catalog record

CREATORNeil D. Lawrence
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE77 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-06795

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

The Inaccessible Game Neil D. Lawrence Abstract In this paper we introduce the inaccessible game , an information-theoretic dynamical system constructed from four axioms. The first three axioms are known and define information loss in the system. The fourth is a novel information isolation axiom that assumes our system is isolated from observation, making it observer-independent and exchangeable. Under this isolation axiom, total marginal entropy is conserved: ∑ i h i = C \sum_{i}h_{i}=C . We consider maximum entropy production in the game and show that the dynamics exhibit a GENERIC-like structure combining reversible and irreversible components. 1 Introduction It is common to think of information as flowing, and information theory ( 30 ) gives us notions of information capacity. More recently, when comparing intelligences, 25 focuses on constraints that dictate how information moves. The book uses the notion of an information topography “In geography, the topography is the configuration of natural and man-made features in the landscape… These questions are framed by the topography. An information topography is similar, but instead of the movement of goods, water and people, it dictates the movement of information.” However, no formal definition of an information topography is given. This leaves the concept as a metaphor without mathematical teeth. This paper is an attempt to address that weakness. In it we formalise a class of information dynamics. We do so by introducing the inaccessible game , an abstract mathematical framework built on three established axioms and one novel postulate. The game builds on ideas from category theory that use three axioms to show that information loss is uniquely measured by (scaled) entropy difference ( 4 ) . We add a fourth postulate of information isolation which, under exchangeability and extensivity requirements, motivates conservation of total marginal entropy. The dynamics of the game emerge from constrained maximum entropy production within the information geometry. We show that these dynamics exhibit a GENERIC-like structure containing both dissipative and conservative elements and relate those dynamics to the physics of our natural world. The inaccessible game follows in a tradition where simple rules can lead to complex emergent phenomena. John Conway’s Game of Life ( 13 ) and Stephen Wolfram’s one-dimensional cellular automata ( 36 ) show that even when dynamics are deterministic zero-player games can exhibit significant complexity that is difficult to predict. 37 called this behaviour computational irreducibility. Our game differs from this tradition in that the rules emerge from the information geometry ( 3 ) . The dynamics are regulated through the Fisher information which operates as a conductance tensor and provides us with our information topography. The rest of this paper is organised as follows. We first review the information loss axioms of 4 which give us our notion of information as represented by entropy (Section 2.1 ). We then introduce the fourth axiom, information isolation , that constrains the system and makes the game inaccessible to an external observer. In Section 3 we introduce an information relaxation principle that governs the dynamics of the game through constrained maximisation of entropy production ( 21 ; 22 ; 23 ; 38 ; 26 ; 7 , MaxEnt,) and show that these dynamics conform to the GENERIC structure, simulating them in Ising model variants in Section 4.3 and 5.2 . We relate our entropy constraint to GENERIC’s more traditional energy constraints in Section 5 . Our final demonstration of the approach (Section 6.2 ) focusses on simple harmonic motion in a mass-spring system. Assumptions and Derivations To aid the reader, we clarify what is axiomatic, what is a modelling restriction, and what is derived. • Axiomatic: Axioms 1–3 (functoriality, convex linearity, continuity) from 4 that uniquely determine information loss as scaled Shannon entropy. • Postulated: Axiom 4 (information isolation) stating the system is isolated from observation. Under exchangeability and extensivity requirements, this implies conservation of total marginal entropy: \sum_{i}h_{i}=C . Additionally, we postulate an information relaxation principle : the system evolves to maximise entropy production subject to the marginal entropy constraint. • Modelling restriction: We restrict to exponential family distributions parametrised by natural parameters \boldsymbol{\theta} . This enables tractable computation but is not yet proven necessary from the axioms alone. • Derived: The specific form of the constrained dynamics, the GENERIC-like decomposition, and the energy-entropy equivalence in the thermodynamic limit all follow from the above postulates and restrictions. 2 Axioms and State Space 30 introduced entropy as a measure of information derived from probability through five axioms. More recently 4 showed that entropy can be derived through category theory as a way of measuring information loss of a measure-preserving function. Without invoking probability, they showed that the Shannon entropy emerges from three axioms for information loss: functoriality, convex linearity, and continuity. From these axioms they derive the information loss as the difference between two Shannon entropies. Of particular interest to us is the second axiom, convex linearity. Functoriality suggests that given a process consisting of two stages, the amount of information lost in the whole process is the sum of the amounts lost at each stage F(f\circ g)=F(f)+F(g), where \circ represents composition. Convex linearity suggests that if we flip a probability- \lambda coin to decide whether to do one process or another, the information lost is \lambda times the information lost by the first process plus (1-\lambda) times the information lost by the second, F(\lambda f\oplus(1-\lambda)g)=\lambda F(f)+(1-\lambda)F(g). The third axiom, continuity , suggests that if we change a process slightly, the information lost changes only slightly, i.e. F(f) is a continuous function of f . The main result of their work is that … there exists a constant c\geq 0 such that for any f:p\rightarrow q F(f)=c(H(p)-H(q)) where F(f) is the information loss in process f:p\rightarrow q and H(\cdot) is the the Shannon entropy measured before and after the process is applied to the system. 2.1 The Fourth Axiom: Information Isolation For the inaccessible game, we introduce a fourth axiom. In an isolated chamber, mass and energy are conserved; our fourth axiom says that our game is similarly isolated from external observation. Under additional requirements of exchangeability and extensivity (see Appendix A ), information isolation implies that the total marginal entropy should be conserved. For any finite sub-group of N variables the sum of marginal entropies, \{h_{i}\}_{i=1}^{N} , sums to a constant, C , \sum_{i=1}^{N}h_{i}=C, (1) or in other words the total amount of information in the system is conserved. The information conservation constraint is imposed in an exchangeable form across the marginal entropies. The form ensures that we can consider any finite partition of variables from a total that could be countably infinite. The exchangeability idea is inspired by, but differs from, the more usual use of exchangeability in non-parametric probability ( 11 ; 2 ; 8 ) . 4 ’s three axioms provide the foundational notion of entropy loss that justifies our use of marginal entropy in our information conservation axiom. In traditional thermodynamics, energy conservation plays the role of an isolated-system constraint. Here, marginal entropy conservation plays an analogous role: it defines a built-in potential within the information geometry. As we show in Section 3 , the curvature of this potential, encoded in the Fisher information, acts as an intrinsic metric governing how

Read the complete text