Data-Driven Prediction of Chaotic Transition in Periapsis Poincaré Maps

This paper presents a novel data-driven methodology using Local and Global Deformation Map-based Dynamic Mode Decomposition (DMD) to approximate nonlinear chaotic transitions in periapsis Poincaré maps via linear operators, enabling efficient trajectory prediction and the design of ballistic transfers to the Moon within the circular restricted three-body problem.

The Gist

The Big Picture: Navigating a Cosmic Storm

Imagine you are trying to fly a spaceship from Earth to the Moon. In the old days, engineers treated space like a calm lake: they calculated a straight line, fired the engine, and hoped for the best. But space isn't a calm lake; it's a turbulent, chaotic storm created by the gravity of the Earth and the Moon pulling on each other.

In this storm, tiny mistakes are huge problems. If you aim your spaceship just a millimeter off, or if your clock is off by a fraction of a second, your trajectory could spiral wildly away from the Moon. This is called chaos. Predicting where you will end up after a few weeks is like trying to predict exactly where a single drop of ink will land in a swirling cup of coffee after you've stirred it a thousand times.

The Problem: The "Map" is Too Complicated

Scientists use a tool called a Poincaré Map to visualize this chaos. Think of this map as a "snapshot" of your spaceship every time it swings closest to the Earth (its "periapsis"). Instead of tracking the whole 3D flight path, you just look at a 2D dot on a graph showing your distance and angle.

The problem is that calculating where that dot moves next requires solving incredibly complex math equations for every single step of the journey. It's like trying to drive a car by manually calculating the physics of every single molecule in the engine. It's accurate, but it takes forever and requires supercomputers.

The Solution: The "Magic Deformation" Trick

This paper introduces a new, data-driven method called Dynamic Mode Decomposition (DMD). Instead of solving the complex physics equations every time, the authors teach a computer to learn the "shape" of the chaos.

Here is the core idea using an analogy:

Imagine you have a sheet of rubber with a grid drawn on it.

1. The Old Way: You try to calculate exactly how every single molecule of rubber stretches and twists based on the wind.
2. The New Way (This Paper): You take a photo of the rubber sheet, stretch it a little bit, take another photo, and then ask a computer: "What is the simple rule that turns Photo A into Photo B?"

Once the computer learns that rule (a "linear operator"), it can predict what the rubber sheet will look like 100 steps later just by applying that rule over and over again. It's like using a stamp to print the future instead of drawing it by hand.

Two Types of "Rubber Sheets"

The authors developed two versions of this trick to handle different situations:

1. LDMD (The Local Microscope)

• How it works: This method zooms in on a tiny, specific patch of the rubber sheet. It samples the data very densely in that small area.
• Best for: Precision. If you need to land a spaceship on a specific spot on the Moon, you use this. It's like using a magnifying glass to see the tiny wrinkles in the fabric. It's very accurate for short trips but only works if you stay in that specific neighborhood.

2. GDMD (The Global Wide-Angle Lens)

• How it works: This method looks at the entire rubber sheet, but it samples the data more sparsely (fewer points, but spread out everywhere).
• Best for: Big picture planning. It captures the "big swirls" of the chaos. It can tell you, "If you start here, you will likely end up in the Moon's gravity well." It's less precise than the microscope but much faster and covers the whole map.

Why This Matters: The "Ballistic Transfer"

The authors tested this by designing a ballistic transfer to the Moon. This is a fuel-efficient way to get to the Moon that uses the natural "tubes" of gravity (called manifolds) rather than burning a lot of fuel.

Think of these gravity tubes like water slides in a water park.

• Old Method: You try to calculate the exact water pressure and friction for every inch of the slide to see if you'll make it to the bottom.
• New Method: You look at a video of the slide, learn the pattern of how the water flows, and then instantly predict: "If I jump in at this specific spot, I will slide all the way to the Moon."

They used their new method to find the exact starting point (the "jump") that would send a spacecraft through the chaotic storm and safely into the Moon's orbit.

The Takeaway

This paper is a breakthrough because it turns a chaotic, unpredictable nightmare into a predictable, linear puzzle.

• Before: "Let's run a supercomputer simulation for 3 days to see if this path works."
• Now: "Let's apply this simple mathematical stamp, and we'll know in milliseconds if it works."

It bridges the gap between the messy reality of space physics and the clean, fast calculations engineers need to design future missions. It proves that even in the most chaotic systems, there is an underlying pattern that data can find, allowing us to navigate the stars with greater speed and confidence.

Technical Summary

1. Problem Statement

Designing low-energy trajectories in multi-body dynamical systems (specifically the Circular Restricted Three-Body Problem, CRTBP) relies heavily on understanding chaotic transport mechanisms, such as invariant manifolds and resonant gravity assists.

• The Challenge: Predicting the evolution of chaotic trajectories is inherently difficult due to sensitive dependence on initial conditions (the "butterfly effect"). Small errors grow exponentially, rendering long-term numerical integration unreliable or computationally expensive.
• The Bottleneck: While Periapsis Poincaré Maps (PPM) effectively reduce the system's dimensionality and visualize transport structures (like lobes and separatrices), they lack an analytical representation. Consequently, trajectory design requires repeated numerical integrations to evaluate state transitions, creating a significant computational bottleneck.
• Limitation of Standard Methods: Standard Dynamic Mode Decomposition (DMD) typically models the evolution of single trajectories. However, chaotic transport is governed by the collective deformation of ensembles of states. Applying standard pointwise DMD near transport boundaries (separatrices) often fails because a single linear operator cannot simultaneously approximate incompatible local dynamics, leading to averaged-out errors that obscure critical geometric structures.

2. Methodology

The authors propose a novel Data-Driven Modeling Framework that uses Dynamic Mode Decomposition (DMD) to approximate the deformation of a set of periapsis points rather than individual trajectories. This approach models the nonlinear chaotic transport using a linear operator, enabling fast prediction via matrix powers.

Two complementary approaches are developed:

A. Local Deformation Map-based DMD (LDMD)

• Objective: High-resolution characterization of local phase-space deformations for precise trajectory targeting.
• Data Strategy: Uses densely sampled data within a small, localized region of the PPM (e.g., a specific range of semi-major axis $a$ and phase angle $\theta$).
• Mechanism: Constructs a discrete map $A$ that tracks the local deformation of a periapsis set over a short time horizon (e.g., 8 periapsis passages).
• Application: Ideal for refining specific trajectory segments and analyzing local sensitivity (Finite-Time Lyapunov Exponents).

B. Global Deformation Map-based DMD (GDMD)

• Objective: Capturing large-scale transport patterns and chaotic scattering across the entire PPM.
• Data Strategy: Uses sparsely sampled data distributed widely across the phase space. It leverages long-term integration (forward and backward) of multiple orbits to build a comprehensive database ($X_{DB}$).
• Mechanism: Constructs a lower-dimensional linear operator from a larger dataset, capturing global deformation structures without requiring pre-selected periodic orbits.
• Application: Suitable for preliminary mission design, identifying global transport channels, and reconstructing trajectories from sparse initial conditions.

Key Technical Innovation:
Instead of modeling $x_{k+1} = Ax_k$ for a single point, the method models the evolution of a vector $x_k$ containing the stacked states of $n$ orbits. The operator $A$ is computed via least-squares ($A = X'X^\dagger$), implicitly enforcing consistency across the neighborhood. This allows the method to capture the "stretching and folding" of phase space sets.

3. Key Results

• High-Accuracy Recovery:
  • LDMD: Achieved extremely low approximation errors ($\epsilon_\theta \approx 10^{-6}$ degrees, $\epsilon_a \approx 10^{-5}$ km) over 8 periapsis steps. The error distribution revealed sharp boundaries corresponding to invariant manifolds, proving the method captures underlying geometric structures.
  • GDMD: Successfully reconstructed orbital trajectories even with sparse data. The method demonstrated that reliable discrete maps can be built with as few as 4 initial orbits in chaotic regions, provided they cover the necessary dynamical behavior.
• Comparison with State Transition Matrix (STM):
  • LDMD significantly outperformed the traditional STM approach in terms of computational cost and accuracy for out-of-sample predictions. STM errors were higher due to time-step misalignment between the reference and perturbed trajectories, whereas LDMD used direct matrix multiplication.
• FTLE and Kick Function Analysis:
  • The method successfully identified Finite-Time Lyapunov Exponent (FTLE) ridges, which correspond to transport barriers, without prior knowledge of periodic orbits.
  • The reconstructed "kick function" (variation in semi-major axis $\delta a$ vs. phase angle $\theta$) matched numerical simulations, confirming that the data-driven models correctly capture resonance-driven chaotic dynamics.
• Trajectory Design Application:
  • The authors successfully designed a ballistic lunar transfer using a targeting strategy. By reversing the discrete map (back-propagation) from a target point within the L1 Lyapunov orbit's stable manifold tube, they identified an initial condition that successfully guided a spacecraft to the Moon.

4. Key Contributions

1. Set-Deformation DMD: A reformulation of DMD that models the deformation of ensembles of states rather than single trajectories, overcoming the limitations of standard DMD near separatrices.
2. Dual-Approach Framework: The introduction of LDMD for local precision and GDMD for global transport analysis, offering flexibility for different mission design phases.
3. Data-Driven Manifold Discovery: Demonstrated that linear operators derived from data can reveal invariant manifold structures and transport barriers without explicit analytical knowledge of the system's periodic orbits.
4. Computational Efficiency: Replaced expensive, repeated numerical integrations with efficient matrix power operations, enabling rapid trajectory prediction and optimization.

5. Significance

This work bridges the gap between chaotic dynamics theory and practical spacecraft trajectory design. By converting complex, nonlinear chaotic transport into a linear, data-driven discrete map, the authors provide a powerful tool for:

• Rapid Mission Design: Enabling fast exploration of low-energy transfer options in the Earth-Moon system and beyond.
• Robustness: Offering stable finite-time predictions even in highly chaotic regions where traditional methods struggle.
• Interpretability: Providing direct geometric insight into phase-space deformation and transport pathways through the spectral properties of the learned linear operators.

The study validates that data-driven modeling is not just a predictive tool but a viable methodology for uncovering the fundamental geometric structures governing multi-body dynamics.