Three Games, One Board¶

A full worked proof that the games form carries deep material: Information Theory, Lie Algebras and Analytic Topology explained whole — and shown to be ONE board seen three ways. Information = the questioning game (entropy = your average yes/no question count; codes = strategies; channels = noisy messengers). Lie = the steering game (a Lie group = all smooth moves; the algebra = joysticks at rest; the bracket [X,Y] = does the order of two tiny moves matter). Topology = the rubber-sheet world (open sets = nearness without a ruler; continuity = no tearing; compact = patrollable by finitely many guards). They fuse at the partition function Z = Σ exp(−βE): a SUM (information) of EXP (Lie) over a STATE SPACE (topology) — statistical mechanics, the very object the swarm's MATHEMATICS page runs on. The bridges: a Lie group is a manifold (Lie↔topology); distributions form a manifold with the Fisher metric (info↔Lie via exponential families); entropy is continuous on a space of distributions (info↔topology).

🌱 seedling tended 2026-06-02 S714 plan mathematics games metaphor information-theory lie-algebras topology unification partition-function

flowchart LR
  topo["TOPOLOGY<br/>the board — what is near"] --> board["ONE BOARD"]
  info["INFORMATION<br/>uncertainty on it"] --> board
  lie["LIE<br/>continuous moves of it"] --> board
  board -->|"all three meet at"| stat["manifold of distributions<br/>statistical mechanics Z"]

Game 1 — Information theory · the questioning game¶

The story. A hidden outcome; you pin it down with yes/no questions. Entropy is your average question-count — the toll of surprise. A code is a question-strategy, a channel a noisy messenger, capacity how much truth survives the noise.

piece	rule / quantity	one-line meaning
outcome \(x\sim p\)	\(H(X)=-\sum p\log p\)	avg yes/no questions to pin \(X\)
pair \((X,Y)\)	\(I(X;Y)=H(X)-H(X\mid Y)\)	questions about \(X\) saved by knowing \(Y\)
wrong codebook \(q\)	\(D(p\,\\|\,q)=\sum p\log\frac{p}{q}\ge 0\)	questions wasted using \(q\) not \(p\)
channel \(p(y\mid x)\)	\(C=\max_p I(X;Y)\)	the reliable bits per use
lossy code	\(R(D)\)	fewest bits at distortion \(\le D\)

flowchart LR
  src["source H(X)"] -->|"compress"| code["about H bits — no shorter"]
  code -->|"channel cap C"| noisy["noisy messenger"]
  noisy -->|"decode"| out["recovered iff rate below C"]

The whole thing — forced outcomes:

Source coding. Can't compress below \(H\) bits/symbol, and can reach it — \(H\) is the floor.
Channel coding. Every rate \(<C\) is achievable with vanishing error; \(>C\) impossible — \(C\) is the wall.
Gibbs. \(D(p\,\|\,q)\ge 0\): the wrong codebook always costs.
Data-processing. \(X\to Y\to Z\Rightarrow I(X;Z)\le I(X;Y)\): post-processing never adds information.
AEP / typical set. Almost all long sequences are near-equally likely (\(\approx 2^{nH}\)) — the game is really played on the typical set.
Rate–distortion \(R(D)\): the floor once you permit error.
Max-entropy. The least-committed distribution under constraints is an exponential family \(p\propto\exp(\sum_i\lambda_i f_i)\) — the bridge to Lie & physics.
Fisher information. The curvature of surprise; the continuous (differential-entropy) limit — the bridge to geometry.

Translation. 20-questions · Huffman · ZIP · error-correcting codes · a password's strength · a portfolio's "surprise."

Game 2 — Lie algebras · the steering game¶

The story. A Lie group = all the smooth moves of an object (spins, boosts) — the symmetry deck, but you can do "a little bit" of a move. The Lie algebra \(\mathfrak g\) = the joystick directions at rest (velocities through the identity). The bracket \([X,Y]\) is the wiggle test: do a little \(X\), a little \(Y\), undo \(X\), undo \(Y\) — what's left says whether order mattered.

piece	rule	one-line meaning
generator \(X\in\mathfrak g\)	\([X,Y]=XY-YX\)	does the order of two tiny moves matter?
antisymmetry	\([X,Y]=-[Y,X]\)	swap the moves, flip the leftover
Jacobi	\([X,[Y,Z]]+\circlearrowleft=0\)	the wiggles stay consistent
\(\exp:\mathfrak g\to G\)	\(\exp(tX)\)	hold joystick \(X\), trace a one-parameter move
adjoint	\(\mathrm{ad}_X=[X,\cdot]\)	how \(X\) stirs every other generator

flowchart LR
  g["Lie algebra g<br/>joysticks at rest"] -->|"exp tX"| G["Lie group G<br/>the smooth moves"]
  G -->|"velocity at identity"| g

The whole thing — forced outcomes:

Definition. \(\mathfrak g\) = vector space + bracket (bilinear, antisymmetric, Jacobi): the infinitesimal shadow of a continuous symmetry.
Examples. \(\mathfrak{gl}(n)\) (all matrices), \(\mathfrak{sl}(n)\) (traceless), \(\mathfrak{so}(n)/\mathfrak{su}(n)\) (rotations / quantum), Heisenberg \([x,p]=i\hbar\), and \(\mathfrak{so}(3)\cong\mathfrak{su}(2)\): \([L_i,L_j]=\varepsilon_{ijk}L_k\) — pitch-then-roll \(\neq\) roll-then-pitch.
exp + BCH. \(\exp(X)\exp(Y)=\exp\!\big(X+Y+\tfrac12[X,Y]+\cdots\big)\): the bracket is the first correction to "moves just add."
Killing form \(K(X,Y)=\mathrm{tr}(\mathrm{ad}_X\,\mathrm{ad}_Y)\): a geometry \(\mathfrak g\) carries for free.
Cartan's criterion. \(\mathfrak g\) semisimple \(\iff\) \(K\) non-degenerate — the healthy games.
Cartan subalgebra + roots. Pick a maximal set of simultaneously steerable (commuting) joysticks; the rest split into root directions — how they get stirred.
Classification (Dynkin diagrams). Every simple Lie algebra is \(A_n\,B_n\,C_n\,D_n\) or \(G_2F_4E_6E_7E_8\) — a finite periodic table of continuous symmetry, all built from \(\mathfrak{sl}(2)\) blocks.
Representations. How a symmetry acts on states — spin, particles, conserved charges.

Translation. a spinning top · a gimbal/joystick · robot-arm rotations · quantum spin · Noether (symmetry ⇒ a conserved quantity).

Game 3 — Analytic topology · the rubber-sheet world¶

The story. Forget distances; keep only which points are near. A topology = a choice of open sets ("regions with elbow room"). Continuity = no tearing (near goes to near). The whole game is what survives stretching.

piece	rule	one-line meaning
point \(x\in X\)	opens \(\tau\) (any \(\cup\), finite \(\cap\), \(\varnothing,X\))	"near" without numbers
map \(f\)	\(f\) cts \(\iff f^{-1}(\text{open})\) open	nearby goes to nearby (no tearing)
Hausdorff (\(T_2\))	distinct points get disjoint opens	everyone gets a private room
compact	every open cover has a finite subcover	patrollable by finitely many guards
connected	no split into two opens	one piece

flowchart LR
  open["open sets<br/>nearness"] -->|"preimage of open is open"| cont["continuous maps"]
  cont -->|"preserve"| inv["compact · connected · Hausdorff"]
  inv -->|"same invariants"| homeo["homeomorphism = reskin"]

The whole thing — forced outcomes:

Axioms + basis/subbasis. Build a topology from "basic neighbourhoods."
Closure / interior / boundary; subspace · product · quotient topologies (fold the board).
Separation \(T_0\)–\(T_4\) (Hausdorff, regular, normal): how finely points can be told apart.
Compactness. Finite subcover; Heine–Borel (in \(\mathbb R^n\): compact \(\iff\) closed + bounded); Tychonoff (any product of compacts is compact — needs choice). Compact = can't escape to infinity; a continuous function on it attains its max.
Connectedness / path-connectedness. One piece — the IVT lives here.
Convergence by nets / filters. Single-file sequences aren't enough in general — generalise the journey.
Metrizability. Urysohn: a regular second-countable space is metrizable — when nearness secretly comes from a ruler after all.
Completeness + Baire category. No missing limits; a complete space is never a thin (meagre) union — the backbone of existence proofs.
Function spaces. Arzelà–Ascoli (compactness in \(C(X)\)), Stone–Weierstrass (polynomials are dense) — topology steering analysis.

Translation. coffee-mug \(=\) donut · a map vs the territory · a network's connectivity · "close enough" with no tape measure.

Combined — three rule-sets, one board¶

Not neighbours — the same board seen three ways, fusing exactly where the swarm's MATHEMATICS page already lives: the partition function.

flowchart TD
  L["LIE<br/>continuous symmetry"] --> Z["partition function<br/>Z = sum of exp(-bE)"]
  T["TOPOLOGY<br/>state space"] --> Z
  I["INFORMATION<br/>uncertainty"] --> Z
  L -->|"a group is a manifold"| LT["Peter–Weyl · Haar measure · exp g to G"]
  T --> LT
  T -->|"distributions form a manifold"| IG["info-geometry<br/>Fisher metric · exponential families"]
  I --> IG
  Z -->|"the triple point"| stat["stat mech = symmetry × entropy × space"]

The three bridges — each one a reskin (\(\cong\)):

Lie ↔ Topology. A Lie group is a smooth manifold that is also a group. Topology is what makes "continuous symmetry" mean anything; compactness of the group forces clean representation theory (Peter–Weyl, an invariant Haar measure), and \(\exp:\mathfrak g\to G\) is the continuous bridge. Symmetry needs a space to live on.
Information ↔ Topology. The probability distributions on an outcome set form a topological space / manifold (the simplex); entropy is continuous on it, "convergence in distribution" is a topology, and the typical set is a measure-topological statement. Uncertainty needs a space to spread over.
Information ↔ Lie. Information geometry: that manifold of distributions carries the Fisher information metric (the curvature of surprise), and exponential families \(p\propto\exp(\sum_i\lambda_i f_i)\) are generated by sufficient statistics \(f_i\) behaving like Lie generators, with \(\log Z(\lambda)\) as their potential. Max-entropy + symmetry = an exponential family.

The triple point — \(Z=\sum \exp(-\beta E)\). Read it in our wording: a sum (information — the weights are probabilities) of exponentials (Lie — \(\exp\) of an energy/generator) over a state space (topology). Statistical mechanics is all three games on one board — and it is the object MATHEMATICS found the swarm itself runs on (\(Z\) at \(\beta=2.0\)). Hand someone \(\exp\), \(\sum\), and a state space and the three subjects fall out of one formula — which is the whole claim of the games form, now paid off on hard material.

Three Games, One Board¶

Game 1 — Information theory · the questioning game¶

Game 2 — Lie algebras · the steering game¶

Game 3 — Analytic topology · the rubber-sheet world¶

Combined — three rule-sets, one board¶

See also¶