pith. sign in

arxiv: 2510.17193 · v2 · pith:HED3ONVRnew · submitted 2025-10-20 · ❄️ cond-mat.dis-nn · cond-mat.stat-mech

Optimal transport by a Lagrangian dynamics of population distribution

Babak Benam , Abolfazl Ramezanpour This is my paper

Pith reviewed 2026-05-21 21:16 UTC · model grok-4.3

classification ❄️ cond-mat.dis-nn cond-mat.stat-mech
keywords human mobilityLagrangian dynamicspopulation distributiondissipationgradient descenturban dynamicsoptimal transportgenerative model
0
0 comments X

The pith

A quadratic Lagrangian with dissipation models human mobility as stationary trajectories of local population functions.

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

The paper develops a Lagrangian dynamical model for movement processes in which local population functions serve as the coordinate variables. Parameters of this Lagrangian are recovered through an efficient gradient descent procedure that minimizes a local error function tied to the observed dynamics. Even the simplest quadratic form that includes dissipation is shown to reproduce the behavior of both synthetic test cases and real empirical movement records. This matters because it supplies an interpretable generative description of how inertia, interactions, and randomness together shape mobility across urban scales.

Core claim

We develop a Lagrangian dynamical model for movement processes, using local population functions as the coordinate variables. An efficient gradient descent algorithm is introduced to estimate the optimal Lagrangian parameters minimizing a local error function of the dynamical process. We show that even a quadratic Lagrangian, incorporating dissipation, effectively captures the dynamics of synthetic and empirical movement data. The inferred models reveal that inertia and dissipation are of comparable importance, while interactions and randomness in the movements induce significant qualitative changes in model parameters. Our results provide an interpretable and generative model for human mobi

What carries the argument

Lagrangian dynamical model whose coordinates are local population functions, with parameters recovered by gradient descent on a local error function

If this is right

  • Inertia and dissipation play roles of comparable importance in the fitted dynamics.
  • Interactions and randomness produce significant qualitative shifts in the recovered model parameters.
  • The fitted Lagrangian supplies an interpretable generative model for human mobility.
  • The same framework supports applications in movement prediction.

Where Pith is reading between the lines

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

  • The same parameter-recovery procedure could be applied to mobility records from different cities or transport modes to compare how interaction strength changes the effective Lagrangian.
  • If the recovered parameters prove unique across repeated fits on similar data, the model could serve as a diagnostic tool for detecting regime shifts in urban flows.
  • Viewing the action as an effective cost functional opens a direct link to classical optimal-transport problems without requiring an explicit cost matrix.

Load-bearing premise

Observed movement processes can be represented as stationary trajectories of an effective action whose coordinates are local population functions, with parameters that can be uniquely recovered by minimizing a local error function via gradient descent.

What would settle it

Empirical movement data for which gradient descent on the quadratic Lagrangian parameters fails to produce stationary trajectories that match the observed population changes within the reported error tolerance.

Figures

Figures reproduced from arXiv: 2510.17193 by Abolfazl Ramezanpour, Babak Benam.

Figure 1
Figure 1. Figure 1: FIG. 1. Learning a Lagrangian dynamics by a local gradient descent algorithm. [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Comparing the synthetic and inferred model dynamics in a movement process of [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Probability distribution of the model parameters (Γ [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Probability distribution of the local ratio of inertia to dissipation ( [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Probability distribution of the local fluxes (Φ [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Color map of the local dynamical susceptibilities to initial populations ( [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Color map of the local dynamical susceptibilities to initial populations ( [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
read the original abstract

Human mobility, enabled by diverse transportation modes, is fundamental to urban functionality. Studying these movements across scales-from microscopic to macroscopic-yields valuable insights into urban dynamics. Local adaptation and (self-)organization in such systems are expected to result in dynamical behaviors that are represented by stationary trajectories of an appropriate effective action. In this study we develop a Lagrangian dynamical model for movement processes, using local population functions as the coordinate variables. An efficient gradient descent algorithm is introduced to estimate the optimal Lagrangian parameters minimizing a local error function of the dynamical process. We show that even a quadratic Lagrangian, incorporating dissipation, effectively captures the dynamics of synthetic and empirical movement data. The inferred models reveal that inertia and dissipation are of comparable importance, while interactions and randomness in the movements induce significant qualitative changes in model parameters. Our results provide an interpretable and generative model for human mobility, with potential applications in movement prediction.

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

3 major / 2 minor

Summary. The manuscript develops a Lagrangian dynamical model for human mobility processes, taking local population functions as coordinate variables. It introduces an efficient gradient-descent procedure to infer the coefficients of a quadratic Lagrangian (including a dissipation term) by minimizing a local error functional on observed trajectories. The central claim is that this quadratic model, once fitted, effectively reproduces both synthetic and empirical movement data, that inertia and dissipation are of comparable magnitude, and that interaction and randomness terms induce qualitative shifts in the recovered parameters. The approach is positioned as providing an interpretable, generative description of mobility with potential predictive uses.

Significance. If the fitting procedure can be shown to yield unique, stable, and predictive parameters rather than merely reproducing the training trajectories, the work would supply a principled, physics-inspired framework for modeling collective movement that bridges microscopic rules and macroscopic flows. The explicit inclusion of dissipation and the use of population densities as dynamical variables distinguish it from standard optimal-transport or diffusion approximations and could enable new falsifiable predictions about long-term mobility statistics.

major comments (3)
  1. [section describing the gradient-descent algorithm and parameter recovery] The manuscript provides no analysis of the loss landscape, Hessian, or multiple minima for the quadratic Lagrangian optimization. Because the loss is quadratic in the coefficients, inertia and dissipation terms can trade off while producing comparable local errors; different initial seeds may therefore converge to qualitatively distinct parameter sets that agree on short-time statistics but diverge on longer horizons. This directly affects the claim that the inferred model “effectively captures” the dynamics and that inertia and dissipation are “of comparable importance.”
  2. [results on synthetic and empirical data] All reported success metrics appear to be computed on the same trajectories used to fit the Lagrangian coefficients. No out-of-sample tests, cross-validation, or comparison against held-out synthetic or empirical data are described, leaving open the possibility that the quadratic form simply overfits the training movements rather than recovering the underlying generative process.
  3. [abstract and empirical-results section] The abstract and results assert that the quadratic Lagrangian “captures the data,” yet no quantitative error values, baseline comparisons (e.g., against pure diffusion, standard OT, or non-dissipative models), or goodness-of-fit statistics are supplied in the provided summary. Without these numbers it is impossible to judge whether the fit is meaningfully better than simpler alternatives or merely consistent with noise.
minor comments (2)
  1. [model-definition section] Notation for the local population functions and the precise definition of the local error functional should be introduced with an explicit equation number at first use to aid reproducibility.
  2. [figures] Figure captions for trajectory comparisons should state the time horizon shown and whether the plotted paths are in-sample or out-of-sample.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their detailed and constructive review of our manuscript. We address each of the major comments below, indicating the revisions we plan to make to improve the clarity and robustness of our results.

read point-by-point responses

  1. Referee: The manuscript provides no analysis of the loss landscape, Hessian, or multiple minima for the quadratic Lagrangian optimization. Because the loss is quadratic in the coefficients, inertia and dissipation terms can trade off while producing comparable local errors; different initial seeds may therefore converge to qualitatively distinct parameter sets that agree on short-time statistics but diverge on longer horizons. This directly affects the claim that the inferred model “effectively captures” the dynamics and that inertia and dissipation are “of comparable importance.”

    Authors: We appreciate the referee pointing out the need for a more thorough examination of the optimization procedure. Although our loss functional is based on local errors along observed trajectories rather than being strictly quadratic in the Lagrangian coefficients, we acknowledge that parameter trade-offs between inertia and dissipation could exist. In the revised manuscript, we will include an analysis of the loss landscape, compute the Hessian at the converged points to assess local convexity, and demonstrate robustness by running the gradient descent from multiple initial seeds. This will support our claims regarding the comparable importance of inertia and dissipation. revision: yes

  2. Referee: All reported success metrics appear to be computed on the same trajectories used to fit the Lagrangian coefficients. No out-of-sample tests, cross-validation, or comparison against held-out synthetic or empirical data are described, leaving open the possibility that the quadratic form simply overfits the training movements rather than recovering the underlying generative process.

    Authors: We agree that out-of-sample validation is essential to rule out overfitting and to confirm that the model captures the underlying dynamics. The current results focus on fitting and reproduction on the available data sets. We will revise the manuscript to include cross-validation procedures, splitting both synthetic and empirical trajectories into training and test sets, and report performance metrics on held-out data. This will provide stronger evidence for the model's predictive capabilities. revision: yes

  3. Referee: The abstract and results assert that the quadratic Lagrangian “captures the data,” yet no quantitative error values, baseline comparisons (e.g., against pure diffusion, standard OT, or non-dissipative models), or goodness-of-fit statistics are supplied in the provided summary. Without these numbers it is impossible to judge whether the fit is meaningfully better than simpler alternatives or merely consistent with noise.

    Authors: We note that the full manuscript does contain some quantitative measures of fit quality, but we accept that more comprehensive reporting and comparisons are necessary for a complete evaluation. In the revision, we will explicitly provide numerical error values (e.g., average trajectory deviation), include direct comparisons to baseline models such as pure diffusion, standard optimal transport, and non-dissipative versions of our Lagrangian, and add goodness-of-fit statistics. These additions will allow readers to better assess the improvement over simpler alternatives. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain.

full rationale

The paper defines a Lagrangian model with population functions as coordinates, introduces a gradient-descent procedure to minimize a local error on observed trajectories, and reports that the resulting quadratic form (with dissipation) reproduces the input synthetic and empirical data. This is standard parameter estimation against external benchmarks rather than a self-referential derivation: the model ansatz is stated independently of the data, the minimization is an explicit fitting step, and success is measured by agreement with the same data used for fitting. No equations reduce to each other by construction, no uniqueness theorem is imported via self-citation, and no fitted quantity is relabeled as an independent prediction. The procedure is therefore self-contained and does not exhibit the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The model rests on one domain assumption and several fitted parameters; no new entities are postulated.

free parameters (1)
  • Lagrangian coefficients
    Coefficients of the quadratic terms for inertia, dissipation, and interactions are adjusted by gradient descent to minimize the local error on movement data.
axioms (1)
  • domain assumption Movement processes are represented by stationary trajectories of an effective action with local population functions as coordinates.
    Invoked at the outset to justify the Lagrangian formulation for human mobility.

pith-pipeline@v0.9.0 · 5682 in / 1256 out tokens · 42767 ms · 2026-05-21T21:16:24.642195+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages · 1 internal anchor

  1. [1]

    A Lagrangian neural network then can be used to represent an observed dynamics that is expected to follow the Euler-Lagrange equations [25, 26]

    use Lagrangian mechanics for identification of migration dynamics. A Lagrangian neural network then can be used to represent an observed dynamics that is expected to follow the Euler-Lagrange equations [25, 26]. Such studies are helpful in providing an optimization formulation of the system dynamics in terms of an action functional. In this study, we intr...

    work page 2022

  2. [2]

    forn= 0, . . . , T−1: –estimate the local gradients 8 ∂ ∂Γab(n) en+1 =−(q a(n+ 1)−˜q a(n+ 1)) ˙qb(n)∆t,(15) ∂ ∂Λab(n) en+1 =−(q a(n+ 1)−˜q a(n+ 1))(q b(n)−µ b(n))∆t,(16) –update the model parameters ∆Γab(n) =−η n ∂ ∂Γab(n) en+1,(17) ∆Λab(n) =−η n ∂ ∂Λab(n) en+1,(18)

    work page

  3. [3]

    In the next section we apply the above algorithm to estimate the model parameters which are best to describe the synthetic and empirical data by a Lagrangian dynamics

    regularization –smooth the parameters ifn= 0: Γab(n)←[Γ ab(n) + Γab(n+ 1)]/2,(19) Λab(n)←[Λ ab(n) + Λab(n+ 1)]/2,(20) otherwise: Γab(n)←[2Γ ab(n) + Γab(n−1) + Γ ab(n+ 1)]/4,(21) Λab(n)←[2Λ ab(n) + Λab(n−1) + Λ ab(n+ 1)]/4,(22) –limit the parameters Γab(n)∈(−Γ max,+Γ max),(23) Λab(n)∈(−Λ max,+Λ max).(24) The learning rateη n is a positive number which can ...

    work page 2000

  4. [4]

    Human mobility: Models and applications

    Barbosa H, Barthelemy M, Ghoshal G, James CR, Lenormand M, Louail T, Menezes R, Ra- masco JJ, Simini F, Tomasini M. Human mobility: Models and applications. Physics Reports. 2018 Mar 6;734:1-74

    work page 2018

  5. [5]

    A review of human mobility: Linking data, models, and real-world applications

    Du Y, Aoki T, Fujiwara N. A review of human mobility: Linking data, models, and real-world applications. Journal of Computational Social Science. 2025 Nov;8(4):90

    work page 2025

  6. [6]

    Human mobility and energy: How do human mobility and energy affect urban policy and planning?

    Dorostkar E, Najarsadeghi M, Molavi M, Zali N. Human mobility and energy: How do human mobility and energy affect urban policy and planning?. Journal of Urban Management. 2023 Dec 1;12(4):413-9

    work page 2023

  7. [7]

    The landscape, trends, challenges, and opportunities of sustainable mobility and transport

    Chen Y, Li C, Wang W, Zhang Y, Chen XM, Gao Z. The landscape, trends, challenges, and opportunities of sustainable mobility and transport. npj Sustainable Mobility and Transport. 2025 Feb 21;2(1):8

    work page 2025

  8. [8]

    Limits of predictability in human mobility

    Song C, Qu Z, Blumm N, Barab´ asi AL. Limits of predictability in human mobility. Science. 2010 Feb 19;327(5968):1018-21

    work page 2010

  9. [9]

    A universal model for mobility and migra- tion patterns

    Simini F, Gonz´ alez MC, Maritan A, Barab´ asi AL. A universal model for mobility and migra- tion patterns. Nature. 2012 Apr 5;484(7392):96-100

    work page 2012

  10. [10]

    Emergence of urban growth patterns from human mobility behavior

    Xu F, Li Y, Jin D, Lu J, Song C. Emergence of urban growth patterns from human mobility behavior. Nature Computational Science. 2021 Dec;1(12):791-800

    work page 2021

  11. [11]

    Urban dynamics through the lens of human mobility

    Xu Y, Olmos LE, Mateo D, Hernando A, Yang X, Gonz´ alez MC. Urban dynamics through the lens of human mobility. Nature computational science. 2023 Jul;3(7):611-20

    work page 2023

  12. [12]

    The spatiotemporal scaling laws of urban population dynamics

    Tan X, Huang B, Batty M, Li W, Wang QR, Zhou Y, Gong P. The spatiotemporal scaling laws of urban population dynamics. Nature Communications. 2025 Mar 24;16(1):2881

    work page 2025

  13. [13]

    Distorted insights from human mobility data

    Gallotti R, Maniscalco D, Barthelemy M, De Domenico M. Distorted insights from human mobility data. Communications Physics. 2024 Dec 24;7(1):421

    work page 2024

  14. [14]

    Returners and explorers dichotomy in human mobility

    Pappalardo L, Simini F, Rinzivillo S, Pedreschi D, Giannotti F, Barab´ asi AL. Returners and explorers dichotomy in human mobility. Nature communications. 2015 Sep 8;6(1):8166. 17

    work page 2015

  15. [15]

    The TimeGeo modeling framework for urban mobility without travel surveys

    Jiang S, Yang Y, Gupta S, Veneziano D, Athavale S, Gonz´ alez MC. The TimeGeo modeling framework for urban mobility without travel surveys. Proceedings of the National Academy of Sciences. 2016 Sep 13;113(37):E5370-8

    work page 2016

  16. [16]

    The universal visitation law of human mobility

    Schl¨ apfer M, Dong L, O’Keeffe K, Santi P, Szell M, Salat H, Anklesaria S, Vazifeh M, Ratti C, West GB. The universal visitation law of human mobility. Nature. 2021 May 27;593(7860):522- 7

    work page 2021

  17. [17]

    Universal predictability of mobility patterns in cities

    Yan XY, Zhao C, Fan Y, Di Z, Wang WX. Universal predictability of mobility patterns in cities. Journal of The Royal Society Interface. 2014 Nov 6;11(100):20140834

    work page 2014

  18. [18]

    Predicting commuter flows in spatial networks using a radiation model based on temporal ranges

    Ren Y, Ercsey-Ravasz M, Wang P, Gonz´ alez MC, Toroczkai Z. Predicting commuter flows in spatial networks using a radiation model based on temporal ranges. Nature communications. 2014 Nov 6;5(1):5347

    work page 2014

  19. [19]

    A generalized radiation model for human mobility: spatial scale, searching direction and trip constraint

    Kang C, Liu Y, Guo D, Qin K. A generalized radiation model for human mobility: spatial scale, searching direction and trip constraint. PloS one. 2015 Nov 24;10(11):e0143500

    work page 2015

  20. [20]

    Modelling and predicting forced migration

    Qi H, Bircan T. Modelling and predicting forced migration. Plos one. 2023 Apr 13;18(4):e0284416

    work page 2023

  21. [21]

    Learning the complexity of urban mobility with deep generative collaboration network

    Li Y, Yuan Y, Ding J, Jin D. Learning the complexity of urban mobility with deep generative collaboration network

    work page

  22. [22]

    Deep activity model: A generative approach for human mobility pattern synthesis

    Liao X, Jiang Q, He BY, Liu Y, Kuai C, Ma J. Deep activity model: A generative approach for human mobility pattern synthesis. arXiv preprint arXiv:2405.17468. 2024 May 24

    work page arXiv 2024

  23. [23]

    Learning to generate synthetic human mobility data: A physics- regularized Gaussian process approach based on multiple kernel learning

    U˘ gurel E, Huang S, Chen C. Learning to generate synthetic human mobility data: A physics- regularized Gaussian process approach based on multiple kernel learning. Transportation Re- search Part B: Methodological. 2024 Nov 1;189:103064

    work page 2024

  24. [24]

    Learning universal human mobility patterns with a foundation model for cross-domain data fusion

    Ma H, Liao X, Liu Y, Jiang Q, Stanford C, Cao S, Ma J. Learning universal human mobility patterns with a foundation model for cross-domain data fusion. Transportation Research Part C: Emerging Technologies. 2025 Nov 1;180:105311

    work page 2025

  25. [25]

    Field theory for recurrent mobility

    Mazzoli M, Molas A, Bassolas A, Lenormand M, Colet P, Ramasco JJ. Field theory for recurrent mobility. Nature communications. 2019 Aug 29;10(1):3895

    work page 2019

  26. [26]

    Human mobility description by physical analogy of electric circuit network based on GPS data

    Zhong Z, Takayasu H, Takayasu M. Human mobility description by physical analogy of electric circuit network based on GPS data. Scientific Reports. 2024 Jun 11;14(1):13380

    work page 2024

  27. [27]

    Coupling of Lagrangian mechanics and physics- informed neural networks for the identification of migration dynamics

    Zakharov K, Kovantsev A, Boukhanovsky A. Coupling of Lagrangian mechanics and physics- informed neural networks for the identification of migration dynamics. Smart Cities. 2025 Mar 18 7;8(2):42

    work page 2025

  28. [28]

    Deep Lagrangian Networks: Using Physics as Model Prior for Deep Learning

    Lutter M, Ritter C, Peters J. Deep lagrangian networks: Using physics as model prior for deep learning. arXiv preprint arXiv:1907.04490. 2019 Jul 10

    work page internal anchor Pith review Pith/arXiv arXiv 1907

  29. [29]

    arXiv preprint arXiv:2003.04630 , year=

    Cranmer M, Greydanus S, Hoyer S, Battaglia P, Spergel D, Ho S. Lagrangian neural networks. arXiv preprint arXiv:2003.04630. 2020 Mar 10

    work page arXiv 2003

  30. [30]

    Li R, Dong L, Zhang J, Wang X, Wang W-X, Di Z and Stanley H E 2017 Nat. Commun. 8 1841

    work page 2017

  31. [31]

    Reducing urban traffic congestion due to localized routing decisions

    Li B, Saad D, Lokhov AY. Reducing urban traffic congestion due to localized routing decisions. Physical Review Research. 2020 Sep;2(3):032059

    work page 2020

  32. [32]

    Yabe, T., Tsubouchi, K., Shimizu, T., Sekimoto, Y., Sezaki, K., Moro, E., & Pentland, A. (2024). YJMob100K: City-Scale and Longitudinal Dataset of Anonymized Human Mo- bility Trajectories [Data set]. In Scientific Data (Vol. 11, Number 1, p. 397). Zenodo. https://doi.org/10.5281/zenodo.10836269

    work page doi:10.5281/zenodo.10836269 2024

  33. [33]

    Claudia Bergroth, Olle J¨ arv, Henrikki Tenkanen, Matti Manninen, & Tuuli Toivonen. (2021). A 24-hour dynamic population distribution dataset based on mobile phone data from Helsinki Metropolitan Area, Finland [Data set]. In Scientific Data (Vol. 9, Number 39, p. 19). Zenodo. https://doi.org/10.5281/zenodo.6106064

    work page doi:10.5281/zenodo.6106064 2021

  34. [34]

    Enriched Traffic Datasets for Madrid

    G´ omez, Iv´ an; Ilarri, Sergio (2025), “Enriched Traffic Datasets for Madrid”, Mendeley Data, V2, doi: 10.17632/697ht4f65b.2

    work page doi:10.17632/697ht4f65b.2 2025

  35. [35]

    Efficiency and irreversibility of movements in a city

    Biazzo I, Ramezanpour A. Efficiency and irreversibility of movements in a city. Scientific reports. 2020 Mar 9;10(1):4334

    work page 2020

  36. [36]

    Entropy production of selfish drivers: impli- cations for efficiency and predictability of movements in a city

    Biazzo I, Nezhadhaghighi MG, Ramezanpour A. Entropy production of selfish drivers: impli- cations for efficiency and predictability of movements in a city. Journal of Physics: Complexity. 2021 Sep 30;2(3):035026

    work page 2021

  37. [37]

    Efficiency of energy-consuming random walkers: Vari- ability in energy helps

    Nezhadhaghighi MG, Ramezanpour A. Efficiency of energy-consuming random walkers: Vari- ability in energy helps. Physical Review E. 2025 Jan;111(1):014301. 19

    work page 2025