Topological flow data analysis for transient flow patterns: a graph-based approach

This paper introduces Topological Flow Data Analysis (TFDA), a graph-based method that converts two-dimensional transient flow patterns into discrete dynamical systems to effectively characterize complex fluid behaviors, such as periodic-to-chaotic transitions and energy-enstrophy relationships, in lid-driven cavity flows.

The Gist

Imagine you are watching a pot of thick soup simmering on a stove. At first, the swirls are simple and predictable. But as the heat turns up, the soup starts to churn, creating tiny whirlpools that merge, split, and dance in chaotic ways. For scientists, understanding this "dance" of fluid is like trying to describe a complex conversation by only listening to the volume of the voices. You know something is happening, but you miss the meaning of the words.

This paper introduces a new way to listen to the soup. Instead of just measuring speed or pressure, the authors use a method called Topological Flow Data Analysis (TFDA). Think of TFDA as a "shape-shifting translator" that turns the messy, swirling soup into a simple, organized family tree.

Here is the breakdown of their discovery using everyday analogies:

1. The Problem: Too Much Noise, Not Enough Structure

In fluid dynamics (the study of moving liquids and gases), we usually get massive amounts of data: numbers for speed, pressure, and temperature at every single point. It's like having a library with millions of books but no catalog. You can see the "vortices" (swirls), but tracking how they change over time is incredibly hard. Traditional methods often reduce this data to abstract math that is hard to interpret, like trying to understand a movie by looking at a spreadsheet of pixel colors.

2. The Solution: The "Flow Family Tree"

The authors developed a way to look at a snapshot of the fluid and ask: "What is the shape of the swirls?"

They treat the fluid like a map.

• The Swirls: Imagine the fluid is a landscape. High-pressure areas are mountains, low-pressure areas are valleys. The fluid flows along the contours of this landscape.
• The "Eddies": These are the little whirlpools in the corners of the container (like the corners of a bathtub).
• The Translation: TFDA takes a snapshot of this landscape and converts it into a tree diagram (called a COT).
  • The trunk of the tree is the main flow.
  • The branches represent the swirls in the corners.
  • The leaves represent the tiny details inside those swirls.

Every time the fluid changes its shape, the tree changes. If two swirls merge, two branches snap together. If a swirl splits, a branch forks. This turns a complex, continuous movie of fluid motion into a simple, discrete story of a tree growing and shrinking.

3. The Experiment: The "Lid-Driven Cavity"

To test this, they used a classic physics experiment: a square box with a lid that slides back and forth, dragging the fluid inside.

• Low Heat (Low Speed): The fluid moves in a calm, repeating loop. The "tree" changes in a simple, predictable cycle, like a clock ticking.
• Medium Heat: The fluid gets a bit jittery. The tree starts to branch out in more complex ways, but it's still somewhat rhythmic.
• High Heat (High Speed): The fluid goes chaotic. The tree is constantly changing, with new branches appearing and disappearing rapidly.

4. What They Discovered

By turning the fluid into these "trees," they found some surprising things:

• Counting the Chaos: They realized that as the fluid gets more chaotic, the number of different "tree shapes" it visits explodes. It's like a musician who starts with a simple scale and eventually plays every possible note in every possible order. They found a "tipping point" (around a specific speed) where the number of tree shapes suddenly jumps, signaling the shift from order to chaos.
• The Energy Connection: They noticed that when the "tree" gets complicated (more branches), the fluid loses more energy (it gets hotter/more turbulent). It's as if the fluid has to work harder to maintain a complex family tree.
• The "Who Influenced Whom" Mystery: This is the coolest part. In the chaotic state, they used a detective technique (called Causal Inference) to ask: Does the swirl in the top-left corner cause the swirl in the bottom-left corner to change, or is it the other way around?
  • The Result: In the chaotic state, the top-left corner seems to be the "boss." It dictates the changes in the bottom-left corner. In the calm state, they were just dancing together in sync. TFDA allowed them to see this hierarchy, which standard math tools missed.

5. Why This Matters

Imagine you are a doctor looking at blood flowing through a heart.

• Old Way: You measure the speed of the blood. If it's fast, you know there's a problem, but you don't know why or where the turbulence started.
• TFDA Way: You look at the "family tree" of the blood flow. You can see exactly where a swirl formed, how it grew, and how it influenced the rest of the heart.

In a nutshell:
This paper is about turning the messy, confusing dance of fluids into a simple, readable story. By translating fluid motion into "trees," the authors can track how order turns into chaos, predict when a system is about to break, and understand which parts of the flow are driving the others. It's like giving a scientist a pair of glasses that turns a blurry, chaotic storm into a clear, structured map.

Technical Summary

1. Problem Statement

Fluid dynamics research often struggles to extract meaningful structural information from complex, high-dimensional time-series data (e.g., velocity, pressure, vorticity). While visualization techniques can highlight features like vortices, identifying a reduced-order model that captures the topological evolution of transient flow patterns remains difficult. Traditional methods like Proper Orthogonal Decomposition (POD) or Dynamic Mode Decomposition (DMD) often produce modes that are mathematically optimal but physically difficult to interpret. Furthermore, existing Topological Data Analysis (TDA) methods have been limited in characterizing the dynamics of transient flows, often focusing only on static pattern identification.

The authors aim to bridge this gap by developing a framework to:

1. Quantitatively track the temporal evolution of 2D flow patterns.
2. Reduce continuous flow dynamics to a discrete dynamical system based on topological equivalence.
3. Analyze causal relationships between local flow structures in complex regimes (periodic, quasi-periodic, and chaotic).

2. Methodology: Topological Flow Data Analysis (TFDA)

The core of the paper is the application of Topological Flow Data Analysis (TFDA), a specialized TDA approach for 2D Hamiltonian vector fields (where the streamfunction $\psi$ acts as the Hamiltonian).

A. Theoretical Foundation

• Structural Stability: The method assumes the flow is "structurally stable," meaning its topological structure is robust against small perturbations.
• COT Representation: The instantaneous streamline pattern is converted into a Partially Cyclically Ordered rooted Tree (COT).
  • Local Orbit Structures: Specific flow features (saddles, centers, separatrices) are mapped to unique symbols (e.g., $b_{\pm\pm}$ for figure-eight orbits, $c_{\pm}$ for boundary saddle connections, $\sigma_{\pm}$ for elliptic centers).
  • Tree Construction: The global connectivity of these local structures forms a tree. The root represents the boundary flow direction, and child nodes represent internal structures.
  • String Representation: The tree is encoded as a unique string (the COT representation), serving as a discrete identifier for the flow's topological state.

B. Implementation Pipeline

1. Data Input: Numerical simulations of the Lid-Driven Cavity Flow (Reynolds numbers $Re = 14,000$ to $16,000$) are generated using the Chebyshev-tau method.
2. Extraction: The psiclone library processes the streamfunction field ($\psi$) to identify critical values (saddles/centers) without explicitly calculating separatrices. It filters out noise by ignoring topological changes smaller than a threshold $\epsilon$.
3. Discretization: The continuous time evolution of the flow is converted into a time series of COT strings.
4. Graph Construction: Transitions between unique COT strings are mapped to a Transition Diagram (a directed graph), where nodes are topological states and edges represent transitions.

C. Causal Inference

To analyze interactions between different regions (corners) of the cavity, the authors employ Convergent Cross Mapping (CCM). This observational causal inference method uses the time-delay embedding of COT states at specific corners to determine if the dynamics of one corner predict the dynamics of another, revealing the direction of influence.

3. Key Contributions

1. Reduced-Order Modeling via Topology: The paper demonstrates that complex fluid dynamics can be reduced to a discrete dynamical system on a graph of topological states, offering high interpretability compared to POD/DMD modes.
2. Quantification of Complexity: The number of distinct topological states ($n_{top}$) is proposed as an order parameter to characterize the transition from periodic to chaotic flow.
3. Period Estimation: The authors show that the period of periodic flows can be accurately estimated using COT time series (via ACF and DTW), matching results from physical observables like kinetic energy.
4. Causal Analysis of Flow Structures: The study introduces a method to detect geometric causality between local flow patterns (e.g., corner eddies) using COT sequences, revealing asymmetries in chaotic regimes that standard sensitivity analysis misses.

4. Key Results

The study focuses on the lid-driven cavity flow across three regimes:

A. Periodic Regime ($Re = 14,000$)

• States: Only 6 distinct topological states are observed.
• Dynamics: The flow follows a simple, dominant cyclic transition (a hexagonal cycle in the transition diagram).
• Physics: Topological changes correlate with energy dissipation. States with complex corner structures (e.g., figure-eight orbits) correspond to energy-decreasing phases, while simpler states correspond to energy-increasing phases.

B. Quasi-Periodic Regime ($Re = 15,500$)

• Complexity: The number of topological states increases significantly. A critical behavior is observed around $Re_c \approx 15,400$, where the number of states scales as a power law: $n_{top} \sim (Re - Re_c)^{0.10}$.
• Dynamics: The transition diagram becomes intricate with multiple cycles and shortcuts.
• Correlation: New topological states emerge specifically around local maxima and minima of enstrophy, linking topology to dissipation rates.

C. Chaotic Regime ($Re = 16,000$)

• Persistence: Despite chaos, "simple" topological states (those appearing at low $Re$) persist with high occurrence rates, supporting the theory that turbulent flows retain memory of low-$Re$ structures.
• Causal Asymmetry: CCM analysis reveals a breakdown of symmetry. In the chaotic regime, the topological changes in the top-left corner ($c_0^+$) causally drive changes in the bottom-left corner ($c_1^+$), whereas this relationship is bidirectional or symmetric in periodic regimes.
• Limitations of Linear Response: Standard linear response analysis (interventional) failed to detect this causal asymmetry, whereas the observational CCM based on TFDA successfully identified it.

5. Significance and Conclusion

This work establishes TFDA as a powerful tool for interpretable fluid dynamics.

• Robustness: Unlike POD, COT representations are robust to noise and numerical artifacts, making them suitable for experimental data (e.g., PIV, echocardiography).
• Interpretability: Every node in the transition graph corresponds to a physically meaningful flow structure (e.g., a specific vortex configuration), unlike abstract POD modes.
• New Insights: The method uncovers hidden causal hierarchies in chaotic flows that are invisible to traditional spectral or sensitivity analyses.

The authors conclude that TFDA provides a rigorous mathematical framework to classify and analyze the global evolution of 2D flows, offering a pathway to understanding complex dynamical behaviors through the lens of topology. Future work may extend this to 3D flows via appropriate 2D projections or fully 3D topological frameworks.