-
Notifications
You must be signed in to change notification settings - Fork 0
Examples Feb2026
This section is inspired by Karpathy’s “open notebook” approach — a living repository of mini-experiments, hands-on guides, and explorations using GEM-LTE.
This guide describes the behavior and underlying mechanism of Manifold-Pinned Dynamics, a deterministic modeling approach for geophysical and atmospheric oscillations (such as ENSO, QBO, and PDO).
The core of this framework, as implemented in the GEM-LTE repository, is the realization that "chaotic" climate signals are actually deterministic responses to lunisolar tidal forcing, constrained by the topological "locking" of fluid manifolds.
The modeling software utilizes a nested trigonometric function to represent the nonlinearity of the system. This is an empirical derivation from Laplace's Tidal Equations (LTE) and the Mathieu/Hill equations, which describe oscillators with periodically varying parameters.
The primary modulation used in the code follows this form:
Where:
-
$M(t)$ : The Latent Forcing Layer. This represents the accumulated phase of the incommensurate lunar cycles (Draconic, Anomalistic, and Tropical) relative to the annual impulse points. -
$k$ : The Wavenumber. This determines the "winding" of the manifold. -
$\epsilon$ : The Coupling Strength. This controls the degree of nonlinearity or "warping" of the manifold. -
$\phi$ : The Phase Offset.
In a linear system, the output directly follows the forcing. However, in a manifold-pinned system, the latent forcing ($M(t)$) acts as a coordinate on a topological surface. The
Manifold-pinning occurs when a high-dimensional dynamical system is "bolted" to a skeleton of periodic orbits at specific spatial or temporal intervals.
The Earth's rotation (Length of Day - LOD) and the annual solar cycle act as a stroboscopic gate. Because the equator is a high-speed anchor for angular momentum, fluid systems like ENSO or the QBO must "reset" or "relaminarize" their phase at these points to maintain global momentum balance.
The software models how different manifolds handle the forcing:
-
ENSO (Nino4): High "topological stiffness." It applies a
$1/X$ wavenumber modulation to cancel long-period drifts. The manifold is "pinned" so tightly that long-wave variations are modulated into high-frequency "oblivion." - AMO/PDO: Lower stiffness. These systems allow for Period Halving (or frequency doubling). A 120-year latent tidal period is "rectified" by the manifold into a 60-year observable cycle.
The GEM-LTE framework distinguishes between systems based on their longitudinal symmetry, which dictates which lunar cycles are "allowed" to drive the manifold.
-
Longitudinally Dependent (
$k \neq 0$ ): Systems like ENSO, which have continental boundaries, must absorb all three lunar cycles (Tropical, Draconic, Anomalistic). This creates the complex, seemingly chaotic "Moire" interference patterns. -
Longitudinally Uniform (
$k = 0$ ): Systems like the QBO (a stratospheric ring) or the Chandler Wobble (axial motion) act as symmetry filters.- The QBO disallows the Tropical cycle (longitudinal dependence) and locks to the Draconic cycle (vertical/zonal).
- The Chandler Wobble locks to the Draconic cycle (radial).
When running the GEM-LTE software, the resulting visualizations help identify these topological states:
- Latent Forcing Layer: The "hidden" accumulation of phase. This is the raw "clock" of the incommensurate tidal cycles.
-
LTE Modulation: The transformation of that latent layer into the observable time series through the
$2\pi$ sinusoid modulation. - Running Windowed Correlation: Areas of high correlation represent periods where the system is successfully locked to the manifold. Dips in correlation (phase slips) represent periods where the system transitions between different branches of the manifold.
In this framework, "Chaos" is an illusion caused by viewing high-dimensional deterministic gear-locking without the proper topological map. The GEM-LTE tool is designed to find the specific
Each example includes:
- Step-by-step commands and code
- Annotations and expected outputs
- “Here’s what I learned” notes
- Tips for modification
See Contributing for how to add your own guided examples.