pith. sign in

arxiv: 2604.22011 · v1 · submitted 2026-04-23 · ⚛️ physics.flu-dyn

Lagrangian Proper Orthogonal Decomposition

Ron Shnapp , Stefano Brizzolara This is my paper

Pith reviewed 2026-05-09 20:28 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords Lagrangian trajectoriesproper orthogonal decompositionturbulencemodal analysisparticle dispersionprincipal component analysissynthetic trajectoriesfluid dynamics
0
0 comments X

The pith

Ten modes capture particle dispersion and curvature in turbulent flows

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Lagrangian Proper Orthogonal Decomposition to represent particle trajectories in turbulence as combinations of a small number of shared modes. Velocity time series from many trajectories are normalized separately and fed into principal component analysis with time points as features. Reconstructions from the leading modes recover single-particle dispersion and curvature statistics when only about ten modes are kept, for paths that last roughly the integral time scale. Acceleration tails need thirty to sixty modes. A reader would care because this low-dimensional description points to a practical way to generate many synthetic trajectories from limited data, which matters for predicting how particles spread and mix in chaotic fluids.

Core claim

The authors establish that LPOD extracts leading modes whose structures and energy content are consistent between direct numerical simulations of homogeneous isotropic turbulence and three-dimensional particle-tracking experiments. Truncated reconstructions, formed by summing selected modes, rescaling fluctuations, and integrating in time, reproduce single-particle dispersion and curvature statistics accurately with roughly ten modes for trajectories of integral-time-scale length, while the tails of acceleration distributions require thirty to sixty modes. Longer trajectories demand progressively more modes. The results indicate a possible route to data-driven generation of synthetic paths.

What carries the argument

Lagrangian Proper Orthogonal Decomposition (LPOD) applies principal component analysis to an ensemble of independently normalized particle velocity time series, treating successive time instances as the feature space.

If this is right

  • Dispersion and curvature statistics are recovered accurately with approximately ten modes for trajectories on the order of the integral time scale.
  • Acceleration distribution tails require a larger set of thirty to sixty modes.
  • Leading modes exhibit similar structures and energy distributions in both simulation and experimental data.
  • Longer trajectories require progressively more modes for accurate reconstruction.
  • The modal basis supports stochastic sampling to generate synthetic particle trajectories.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same modal approach could be tested in inhomogeneous or wall-bounded flows to check whether the reported mode counts remain sufficient.
  • Coupling LPOD modes with Eulerian field decompositions might yield reduced-order models for particle-fluid coupling.
  • Stochastic sampling of the extracted modes could produce large ensembles for studying rare dispersion events without full simulations.
  • The preprocessing normalization step may prove useful for other trajectory-based analyses that seek low-dimensional representations.

Load-bearing premise

Independent normalization of each trajectory's velocity time series isolates fluctuations without introducing artifacts that distort the principal components or the integrated reconstructions.

What would settle it

A new ensemble of integral-time-scale trajectories where ten-mode reconstructions deviate measurably from the observed dispersion and curvature statistics, or where sixty-mode reconstructions still miss the acceleration tails, would falsify the accuracy claims.

Figures

Figures reproduced from arXiv: 2604.22011 by Ron Shnapp, Stefano Brizzolara.

Figure 1
Figure 1. Figure 1: (a) First 10 LPOD modes: solid lines denote DNS data and dashed lines 3D-PTV data. (b) Example DNS trajectory reconstruction using an increasing number of modes (2, 4, 8, and 16); the dashed line shows the ballistic reconstruction from the mean velocity, and the black line the original trajectory. (c) Energy distribution normalized such that Í =1 = 100; solid and dashed lines correspond to DNS and 3D-PTV d… view at source ↗
Figure 2
Figure 2. Figure 2: (a) Single particle dispersion; the inset shows the data normalized by the ballistic scaling. (b) Deviation from ballistic dispersion; the inset shows the relative reconstruction error () [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) Second-order Lagrangian longitudinal structure function computed from the DNS dataset. The inset shows the normalized scaling. Orange circles denote the original data, whereas blue diamonds are obtained from the first 16 modes. (b) Probability density function of the velocity increments from the DNS dataset. The time lag varies from 0.1 to 30 . The reconstructed data (blue solid lines) are obtained usi… view at source ↗
Figure 4
Figure 4. Figure 4: (a) Probability density function of the trajectory acceleration. The orange line denotes the original signal, while the blue lines correspond to reconstructions using 4, 8, 16, and 32 modes. The inset shows the kurtosis K of the reconstructed signal normalized with the original signal Kurtosis, with dashed and solid lines indicating experimental and DNS data, respectively. (b) Probability density function … view at source ↗
read the original abstract

We introduce a modal representation for Lagrangian trajectories in turbulence, termed Lagrangian Proper Orthogonal Decomposition (LPOD). An ensemble of particle trajectories is used to construct velocity time series, which are normalized independently for each trajectory to isolate fluctuations. Principal Component Analysis is then applied to the resulting dataset, with temporal instances defining the feature space. The method is tested on trajectories from both direct numerical simulations of homogeneous isotropic turbulence and three-dimensional particle-tracking experiments, showing that the leading modes exhibit similar structures and energy distributions in both cases. Truncated reconstructions are obtained by combining modes and coefficients, rescaling the fluctuations, and integrating in time. For trajectories of the order of the integral time scale, single-particle dispersion and curvature statistics are accurately reproduced using a limited number of modes (c.a. 10), whereas capturing the tails of acceleration distributions requires a larger set (c.a. 30-60). Longer trajectories require progressively more modes for accurate reconstruction. These results suggest a possible route to data-driven generation of synthetic particle trajectories via stochastic sampling of the modal Lagrangian dynamics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces Lagrangian Proper Orthogonal Decomposition (LPOD) as a modal representation for particle trajectories in turbulence. An ensemble of velocity time series is normalized independently per trajectory (subtract mean, divide by rms) to isolate fluctuations; PCA is then applied across the ensemble with time instants as features. Leading modes are shown to exhibit similar structures and energy distributions between DNS of homogeneous isotropic turbulence and 3D particle-tracking experiments. Truncated reconstructions are formed by combining modes and coefficients, rescaling by the original per-trajectory rms, and integrating in time. The central claim is that, for trajectories of length comparable to the integral time scale, single-particle dispersion and curvature statistics are reproduced accurately with approximately 10 modes, while the tails of acceleration distributions require 30-60 modes; longer trajectories need progressively more modes. The approach is suggested as a route to data-driven synthetic trajectory generation.

Significance. If the claims hold, LPOD supplies a simple, data-driven low-dimensional basis for Lagrangian statistics that works across simulation and experiment. The direct comparison of mode structures between DNS and PTV data is a clear strength, as is the demonstration that a modest number of modes can recover lower-order integrals (dispersion, curvature) while higher-order quantities (acceleration tails) require more. The method is reproducible in principle and parameter-light once the normalization is fixed, though its robustness hinges on the normalization step not distorting the very features needed for extreme events.

major comments (2)
  1. [Method section (normalization and PCA procedure)] The independent per-trajectory normalization (subtract mean, divide by own rms) before PCA is load-bearing for the acceleration-tail claim. Because acceleration is the second derivative, segments with extreme curvature may have normalized profiles that differ systematically from the ensemble-average modes; the paper must demonstrate that the reconstructed acceleration PDF tails remain insensitive to this choice or provide a quantitative test (e.g., comparison with un-normalized PCA or amplitude-preserving variants) showing that the 30-60 mode count still recovers the tails.
  2. [Results (reconstruction statistics)] The abstract and results state that dispersion/curvature are 'accurately reproduced' with ~10 modes and acceleration tails with 30-60 modes, yet no explicit error metric, tolerance, or uncertainty quantification is supplied. For instance, what relative L2 error on the dispersion curve or Kolmogorov-Smirnov distance on the acceleration PDF is considered acceptable? Without these, the quantitative mode-count claims cannot be evaluated.
minor comments (2)
  1. [Abstract] The abbreviations 'c.a.' should be replaced by 'approximately' or 'ca.' for standard journal style.
  2. [Reconstruction procedure] Clarify the numerical scheme used for time integration of the reconstructed velocity to obtain position and acceleration, as second-derivative sensitivity makes the integrator choice relevant.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. We address each major comment below and indicate the changes that will be made in the revised manuscript.

read point-by-point responses

  1. Referee: [Method section (normalization and PCA procedure)] The independent per-trajectory normalization (subtract mean, divide by own rms) before PCA is load-bearing for the acceleration-tail claim. Because acceleration is the second derivative, segments with extreme curvature may have normalized profiles that differ systematically from the ensemble-average modes; the paper must demonstrate that the reconstructed acceleration PDF tails remain insensitive to this choice or provide a quantitative test (e.g., comparison with un-normalized PCA or amplitude-preserving variants) showing that the 30-60 mode count still recovers the tails.

    Authors: We agree that the per-trajectory normalization is central to isolating fluctuation shapes and that its effect on second-derivative statistics requires explicit verification. In the revised manuscript we will add a direct comparison of acceleration PDF tails reconstructed from the current per-trajectory normalization against those obtained with a single global rms normalization applied to the entire ensemble. This test will quantify any change in the number of modes needed to recover the tails and thereby demonstrate robustness of the 30-60 mode claim. revision: yes

  2. Referee: [Results (reconstruction statistics)] The abstract and results state that dispersion/curvature are 'accurately reproduced' with ~10 modes and acceleration tails with 30-60 modes, yet no explicit error metric, tolerance, or uncertainty quantification is supplied. For instance, what relative L2 error on the dispersion curve or Kolmogorov-Smirnov distance on the acceleration PDF is considered acceptable? Without these, the quantitative mode-count claims cannot be evaluated.

    Authors: The referee is correct that the manuscript currently lacks explicit quantitative error metrics to support the stated mode counts. In the revision we will define and report concrete metrics: the relative L2 error on the dispersion and curvature curves, and the Kolmogorov-Smirnov distance on the acceleration PDF. We will also state the tolerance thresholds (for example, relative L2 error below 5 % or KS distance below 0.05) at which we consider the statistics accurately reproduced, allowing the mode-count claims to be evaluated objectively. revision: yes

Circularity Check

0 steps flagged

No circularity: LPOD applies standard PCA after fixed normalization with external validation of statistics

full rationale

The paper defines LPOD as independent per-trajectory velocity normalization (subtract mean, divide by RMS) followed by PCA treating time instants as features, then forms reconstructions by modal sum, rescaling, and integration. Dispersion, curvature, and acceleration statistics are computed separately on original trajectories and on these reconstructions, with no equations or steps that define the reported statistics in terms of parameters fitted to the same reconstructed quantities. No self-citations, uniqueness theorems, or ansatzes are invoked to close the chain; the procedure is a direct, unsupervised decomposition whose accuracy claims rest on direct comparison to held-out data features rather than on any reduction to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that per-trajectory normalization isolates fluctuations without loss of dynamical information and that the resulting PCA modes generalize across the tested DNS and experimental data sets.

axioms (1)
  • domain assumption Independent normalization of each trajectory's velocity time series isolates fluctuations without introducing reconstruction bias
    Invoked in the description of the data-preparation step prior to PCA

pith-pipeline@v0.9.0 · 5474 in / 1281 out tokens · 27153 ms · 2026-05-09T20:28:59.995196+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages

  1. [1]

    Karhunen, Kari , journal=

    work page

  2. [2]

    Nature Communications , volume=

    Universal alignment in turbulent pair dispersion , author=. Nature Communications , volume=. 2023 , publisher=

    work page 2023

  3. [3]

    Atmospheric turbulence and radio wave propagation , pages=

    The structure of inhomogeneous turbulent flows , author=. Atmospheric turbulence and radio wave propagation , pages=. 1967 , publisher=

    work page 1967

  4. [4]

    Current science , pages=

    An introduction to the proper orthogonal decomposition , author=. Current science , pages=. 2000 , publisher=

    work page 2000

  5. [5]

    Journal of Fluid Mechanics , volume=

    A hierarchy of low-dimensional models for the transient and post-transient cylinder wake , author=. Journal of Fluid Mechanics , volume=. 2003 , publisher=

    work page 2003

  6. [6]

    Journal of Atmospheric and Oceanic Technology , volume=

    Wave and turbulence separation using dynamic mode decomposition , author=. Journal of Atmospheric and Oceanic Technology , volume=. 2025 , publisher=

    work page 2025

  7. [7]

    Annual review of fluid mechanics , volume=

    The proper orthogonal decomposition in the analysis of turbulent flows , author=. Annual review of fluid mechanics , volume=. 1993 , publisher=

    work page 1993

  8. [8]

    Physics of Fluids , volume=

    Characterizing Lagrangian particle dynamics in decaying homogeneous isotropic turbulence using proper orthogonal decomposition , author=. Physics of Fluids , volume=. 2022 , publisher=

    work page 2022

  9. [9]

    2024 , publisher=

    Characterization of Two-phase Turbulent Flows , author=. 2024 , publisher=

    work page 2024

  10. [10]

    Physical Review E , volume=

    Principal-component analysis of particle motion , author=. Physical Review E , volume=. 2015 , publisher=

    work page 2015

  11. [11]

    Nature Machine Intelligence , volume=

    Synthetic Lagrangian turbulence by generative diffusion models , author=. Nature Machine Intelligence , volume=. 2024 , publisher=

    work page 2024

  12. [12]

    International Journal of Multiphase Flow , volume=

    Generative diffusion models for synthetic trajectories of heavy and light particles in turbulence , author=. International Journal of Multiphase Flow , volume=. 2024 , publisher=

    work page 2024

  13. [13]

    European Journal of Mechanics-B/Fluids , volume=

    Deterministic diffusion models for Lagrangian turbulence: Robustness and encoding of extreme events , author=. European Journal of Mechanics-B/Fluids , volume=. 2026 , publisher=

    work page 2026

  14. [14]

    Communications Physics , volume=

    Stochastic reconstruction of gappy Lagrangian turbulent signals by conditional diffusion models , author=. Communications Physics , volume=. 2025 , publisher=

    work page 2025

  15. [15]

    Proceedings of the National Academy of Sciences , volume=

    Data-driven Mori--Zwanzig modeling of Lagrangian particle dynamics in turbulent flows , author=. Proceedings of the National Academy of Sciences , volume=. 2026 , publisher=

    work page 2026

  16. [16]

    Proceedings of the london mathematical society , volume=

    Diffusion by continuous movements , author=. Proceedings of the london mathematical society , volume=. 1922 , publisher=

    work page 1922

  17. [17]

    Turbulence and the dynamics of coherent structures. I. Coherent structures , author=. Quarterly of applied mathematics , volume=

    work page

  18. [18]

    AIAA aviation 2019 forum , pages=

    A tutorial on the proper orthogonal decomposition , author=. AIAA aviation 2019 forum , pages=

    work page 2019

  19. [19]

    Journal of Fluid Mechanics , volume=

    Lagrangian approach for modal analysis of fluid flows , author=. Journal of Fluid Mechanics , volume=. 2021 , publisher=

    work page 2021

  20. [20]

    Journal of Applied and Computational Mechanics , pages=

    Proper Orthogonal Decomposition Analysis of Turbulent Energy Spectra within a Lagrangian Particle Tracking Framework , author=. Journal of Applied and Computational Mechanics , pages=. 2025 , publisher=

    work page 2025

  21. [21]

    Journal of Turbulence , number=

    A public turbulence database cluster and applications to study Lagrangian evolution of velocity increments in turbulence , author=. Journal of Turbulence , number=. 2008 , publisher=

    work page 2008

  22. [22]

    Physical review letters , volume=

    Curvature of Lagrangian trajectories in turbulence , author=. Physical review letters , volume=. 2007 , publisher=

    work page 2007

  23. [23]

    2019 , publisher=

    Lagrangian Experimental Investigation of Dispersion in Canopy Turbulence , author=. 2019 , publisher=

    work page 2019

  24. [24]

    Physical Review Letters , volume=

    Measurement of Lagrangian velocity in fully developed turbulence , author=. Physical Review Letters , volume=. 2001 , publisher=

    work page 2001

  25. [25]

    Nature , volume=

    Fluid particle accelerations in fully developed turbulence , author=. Nature , volume=. 2001 , publisher=

    work page 2001

  26. [26]

    Annual review of fluid mechanics , volume=

    Turbulent relative dispersion , author=. Annual review of fluid mechanics , volume=. 2001 , publisher=

    work page 2001

  27. [27]

    Annual review of fluid mechanics , volume=

    Lagrangian investigations of turbulence , author=. Annual review of fluid mechanics , volume=. 2002 , publisher=

    work page 2002

  28. [28]

    George , title =

    William K. George , title =. International Seminar on Wall Turbulence, Dubrovnik , year =

    work page

  29. [29]

    Journal of Engineering Mathematics , volume=

    On the discrepancies between POD and Fourier modes on aperiodic domains , author=. Journal of Engineering Mathematics , volume=. 2024 , publisher=

    work page 2024

  30. [30]

    Physical Review Fluids , volume=

    Turbulence-obstacle interactions in the Lagrangian framework: applications for stochastic modeling in canopy flows , author=. Physical Review Fluids , volume=. 2020 , publisher=

    work page 2020

  31. [31]

    Journal of fluid mechanics , volume=

    Criteria for the selection of stochastic models of particle trajectories in turbulent flows , author=. Journal of fluid mechanics , volume=. 1987 , publisher=

    work page 1987

  32. [32]

    Physics of Fluids A: Fluid Dynamics , volume=

    The velocity-dissipation probability density function model for turbulent flows , author=. Physics of Fluids A: Fluid Dynamics , volume=. 1990 , publisher=

    work page 1990

  33. [33]

    Physics of Fluids , volume=

    Stochastic equations with multifractal random increments for modeling turbulent dispersion , author=. Physics of Fluids , volume=. 1994 , publisher=

    work page 1994

  34. [34]

    Physics of Fluids A: Fluid Dynamics , volume=

    Reynolds number effects in Lagrangian stochastic models of turbulent dispersion , author=. Physics of Fluids A: Fluid Dynamics , volume=. 1991 , publisher=

    work page 1991

  35. [35]

    Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography , volume=

    Lagrangian simulation of wind transport in the urban environment , author=. Quarterly Journal of the Royal Meteorological Society: A journal of the atmospheric sciences, applied meteorology and physical oceanography , volume=. 2009 , publisher=

    work page 2009

  36. [36]

    2013 , publisher=

    Statistical fluid mechanics, volume II: mechanics of turbulence , author=. 2013 , publisher=

    work page 2013

  37. [37]

    Journal of Fluid Mechanics , volume=

    Lagrangian statistics from direct numerical simulations of isotropic turbulence , author=. Journal of Fluid Mechanics , volume=. 1989 , publisher=

    work page 1989

  38. [38]

    Photogrammetric determination of particle coordinates , author=

    Particle tracking velocimetry in three-dimensional flows: Part 1. Photogrammetric determination of particle coordinates , author=. Experiments in fluids , volume=. 1993 , publisher=

    work page 1993

  39. [39]

    Experiments in fluids , volume=

    Particle tracking velocimetry in three-dimensional flows: Part II: Particle tracking , author=. Experiments in fluids , volume=. 1993 , publisher=

    work page 1993

  40. [40]

    Journal of Open Source Software , volume=

    MyPTV: a Python package for 3D particle tracking , author=. Journal of Open Source Software , volume=

    work page

  41. [41]

    Physical review letters , volume=

    Lagrangian velocity statistics in turbulent flows: Effects of dissipation , author=. Physical review letters , volume=. 2003 , publisher=

    work page 2003

  42. [42]

    Physics of Fluids , volume=

    Lagrangian structure functions in turbulence: A quantitative comparison between experiment and direct numerical simulation , author=. Physics of Fluids , volume=. 2008 , publisher=

    work page 2008

  43. [43]

    Journal of Fluid Mechanics , volume=

    Inertial range Eulerian and Lagrangian statistics from numerical simulations of isotropic turbulence , author=. Journal of Fluid Mechanics , volume=. 2010 , publisher=

    work page 2010

  44. [44]

    Journal of Turbulence , number=

    Lagrangian statistics in fully developed turbulence , author=. Journal of Turbulence , number=. 2006 , publisher=

    work page 2006

  45. [45]

    Annual review of fluid mechanics , volume=

    Lagrangian properties of particles in turbulence , author=. Annual review of fluid mechanics , volume=. 2009 , publisher=

    work page 2009

  46. [46]

    Science , volume=

    The role of pair dispersion in turbulent flow , author=. Science , volume=. 2006 , publisher=

    work page 2006

  47. [47]

    Physical Review Letters , volume=

    Generalization of turbulent pair dispersion to large initial separations , author=. Physical Review Letters , volume=. 2018 , publisher=

    work page 2018

  48. [48]

    Physical Review Letters , volume=

    Universality and intermittency of pair dispersion in turbulence , author=. Physical Review Letters , volume=. 2022 , publisher=

    work page 2022

  49. [49]

    Physics of Fluids , volume=

    Lagrangian statistics of particle pairs in homogeneous isotropic turbulence , author=. Physics of Fluids , volume=. 2005 , publisher=

    work page 2005

  50. [50]

    Physics of Fluids , volume=

    On the Lagrangian nature of the turbulence energy cascade , author=. Physics of Fluids , volume=. 1994 , publisher=

    work page 1994

  51. [51]

    EPL (Europhysics Letters) , volume=

    The Lagrangian view of energy transfer in turbulent flow , author=. EPL (Europhysics Letters) , volume=

    work page

  52. [52]

    Physical Review Letters , volume=

    Pair Dispersion of Bubbles in Isotropic Turbulence , author=. Physical Review Letters , volume=. 2025 , publisher=

    work page 2025

  53. [53]

    Journal of Fluid Mechanics , volume=

    Turbulent pair dispersion of inertial particles , author=. Journal of Fluid Mechanics , volume=. 2010 , publisher=

    work page 2010