{
 "artifacts": null,
 "category": "dynamics",
 "description": "Train a recurrent or state-space world model as a semiconjugate factorization: the high-dimensional state dynamics \\(G\\) must project through an encoder \\(E\\) to a lower-dimensional latent map \\(F\\), satisfying \\(E\\circ G\\approx F\\circ E\\). Add a connected-fiber regularizer so states with the same latent code form geometrically coherent sets, allowing the model to discard redundant microscopic motion without discarding the entropy-carrying macroscopic dynamics.",
 "formulas_latex": [
  "$$\\pi\\circ g=f\\circ\\pi,\\qquad \\pi(g(x))=f(\\pi(x)).$$",
  "$$h_{top}(g)=h_{top}(f),\\qquad \\pi_*\\mu=\\mu_{\\mathrm{MME}}.$$",
  "$$\\pi^{-1}(z)=\\bigcap_{k=1}^{\\infty}D_k(z),\\qquad D_{k+1}(z)\\subseteq D_k(z).$$",
  "$$\\mathcal L=\\mathcal L_{pred}+\\lambda_{sc}\\mathbb E_x\\left[\\left\\|E_\\psi(G_\\theta(x))-F_\\phi(E_\\psi(x))\\right\\|_2^2\\right]+\\lambda_{fib}\\mathbb E_{x,x',\\alpha}\\left[\\left\\|E_\\psi(\\alpha x+(1-\\alpha)x')-E_\\psi(x)\\right\\|_2^2\\right]+\\lambda_h\\left|\\widehat h_T(G_\\theta)-\\widehat h_T(F_\\phi\\circ E_\\psi)\\right|.$$"
 ],
 "id": 3102,
 "implementation": "1. Integration point: use an encoder \\(E_\\psi\\), a latent recurrent or state-space transition \\(F_\\phi\\), and a decoder \\(D_\\omega\\) inside an existing world model. The full-state transition \\(G_\\theta\\) can be the observed environment transition, a separately learned high-dimensional model, or the model's recurrent update. Train the latent transition through the semiconjugacy residual rather than relying only on decoded next-state error. 2. Pseudocode: sample trajectories \\(x_t\\); compute \\(z_t=E_\\psi(x_t)\\), \\(\\tilde z_{t+1}=F_\\phi(z_t)\\), and \\(\\hat x_{t+1}=D_\\omega(\\tilde z_{t+1})\\). Minimize reconstruction and prediction losses plus \\(\\lambda_{sc}\\|E_\\psi(x_{t+1})-\\tilde z_{t+1}\\|^2\\). Select pairs \\(x,x'\\) with nearby latent codes, draw \\(\\alpha\\sim U[0,1]\\), and penalize encoder variation along their interpolation. Estimate \\(\\widehat h_T\\) for full and latent trajectories using the same fixed-resolution delay-coordinate or k-nearest-neighbor entropy-rate estimator, then add the entropy mismatch penalty. 3. Taken directly from the paper: the semiconjugacy identity, entropy equality, and nested-fiber principle. Estimated empirically: entropy rates, fiber connectedness, and residual magnitudes. 4. First cheap experiment: use a Hénon map embedded into 20 dimensions with nuisance oscillators and compare a standard autoencoding RNN against this model at equal latent dimension. Measure one-, ten-, and one-hundred-step prediction error, latent and full entropy rates, semiconjugacy residual, and fiber diameter after repeated refinement. The quantitative prediction is that latent entropy remains within 10-20% of full entropy while median fiber diameter decreases under refinement. If the residual is sufficiently small, long-horizon error should be governed by the latent dynamical complexity rather than nuisance dimensions. A falsification is a sharp entropy drop or fragmentation of interpolation fibers into disconnected clusters.",
 "math_summary": "The paper uses a semiconjugacy \\(\\pi\\) between maps \\(g\\) and \\(f\\), meaning \\(\\pi(g(x))=f(\\pi(x))\\) for every state \\(x\\). It assumes equal topological entropies \\(h_{top}(g)=h_{top}(f)\\), where \\(h_{top}\\) measures orbit-complexity growth. It establishes that each fiber \\(\\pi^{-1}(z)\\) is connected and can be written as \\(\\bigcap_{k\\geq 1}D_k(z)\\), with closed disks \\(D_{k+1}(z)\\subseteq D_k(z)\\). The neural adaptation replaces \\(g\\) by a learned full-state transition \\(G_\\theta\\), \\(f\\) by a latent transition \\(F_\\phi\\), and \\(\\pi\\) by an encoder \\(E_\\psi\\). The semiconjugacy residual is \\(r(x)=E_\\psi(G_\\theta(x))-F_\\phi(E_\\psi(x))\\). The entropy-preservation target is \\(|\\widehat h_T(G_\\theta)-\\widehat h_T(F_\\phi\\circ E_\\psi)|\\leq\\delta_h\\), where \\(\\widehat h_T\\) is an empirical finite-horizon entropy-rate estimator and \\(\\delta_h\\) is a chosen tolerance. Connected fibers are approximated by requiring interpolations between states with similar latent codes to remain close to that code.",
 "math_tags": [
  "dynamical-systems",
  "information-theory",
  "statistics",
  "statistical-mechanics"
 ],
 "ml_areas": [
  "world-model",
  "rnn",
  "ssm"
 ],
 "paper": {
  "arxiv_id": "2609.03390",
  "arxiv_url": "https://arxiv.org/abs/2609.03390",
  "summary_what_math_gives_to_ml": "The paper provides a concrete semiconjugacy mechanism: when a smooth surface map has the same topological entropy as a pseudo-Anosov reference map in the same isotopy class, its dynamics factor through the reference map via a continuous map satisfying \\(\\pi\\circ g=f\\circ\\pi\\). The nontrivial structural result is that every fiber \\(\\pi^{-1}(x)\\) is connected, represented as a nested intersection of closed topological disks, and that the maximal-entropy measure of the reference system has a unique invariant lift which is metrically isomorphic to it. A transferable neural-network version is a latent-dynamics architecture that separates entropy-carrying macroscopic dynamics from connected, dynamically redundant fibers.",
  "title": "Entropy and semiconjugacy on surfaces",
  "year": "2026"
 },
 "ratings": {
  "difficulty": 6,
  "novelty": 7,
  "usefulness": 6
 },
 "solves": [
  "stability",
  "accuracy",
  "sample-efficiency"
 ],
 "title": "Entropy-Preserving Semiconjugate World Model",
 "url": "https://synthcore.org/idea/3102/entropy-preserving-semiconjugate-world-model",
 "verification": {
  "peer_reviewed": false,
  "status": "unverified",
  "status_label": "Unverified",
  "verdict_source": "deterministic test code (paired-seed permutation statistics)",
  "verification_axes": {
   "benchmark_mechanism": {
    "confirmed": null,
    "tested": false
   },
   "practical_benchmark": {
    "beats_baseline": null,
    "tested": false
   },
   "toy_mechanism_gate": {
    "confirmed": null,
    "tested": false
   }
  }
 }
}